
import pandas as pd
import numpy as np
import matplotlib.pyplot as pltl
import seaborn as sns
from collections import Counter
#from pandas_profiling import ProfileReport
from sklearn.model_selection import train_test_split, GridSearchCV, RepeatedStratifiedKFold
from sklearn.preprocessing import StandardScaler, MinMaxScaler, OrdinalEncoder, OneHotEncoder
from sklearn.feature_selection import SelectKBest, chi2, RFECV
from scipy.stats import chi2_contingency
from sklearn.decomposition import PCA
from sklearn.pipeline import Pipeline

from sklearn.dummy import DummyClassifier
from sklearn.metrics import accuracy_score, confusion_matrix, f1_score, recall_score
from sklearn.ensemble import RandomForestClassifier
from xgboost import XGBClassifier
from sklearn.linear_model import LogisticRegression
from sklearn.tree import DecisionTreeClassifier

from imblearn.combine import SMOTETomek, SMOTEENN
from imblearn.over_sampling import RandomOverSampler, SMOTE, BorderlineSMOTE, SVMSMOTE, ADASYN
from imblearn.under_sampling import RandomUnderSampler, NearMiss, CondensedNearestNeighbour, TomekLinks
from imblearn.under_sampling import EditedNearestNeighbours, OneSidedSelection, NeighbourhoodCleaningRule

import dill

%matplotlib inline


#from google.colab import drive
#drive.mount('/content/drive')


usuariosDF = pd.read_csv('/Users/inigoab/Documents/DS_UOC/TFM/Potenciales_Compradores_ecommerce/Linea1-Usuarios.txt',
    sep=";", decimal=",", encoding = "ISO-8859-1")

#usuariosDF = pd.read_csv("./drive/MyDrive/Potenciales Compradores ecommerce/Linea1-Usuarios.txt", 
#                         sep=";", decimal=",", encoding = "ISO-8859-1")


usuariosDF.info()

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 367705 entries, 0 to 367704
Data columns (total 19 columns):
 #   Column            Non-Null Count   Dtype  
---  ------            --------------   -----  
 0   IDUSUARIO         367705 non-null  float64
 1   TIPOUSUARIO       367705 non-null  object 
 2   FEC_REGISTRO      367705 non-null  object 
 3   CANAL_REGISTRO    360171 non-null  float64
 4   IND_CLIENTE       367705 non-null  int64  
 5   IND_ALTA          367705 non-null  int64  
 6   TIPOEMAIL         367705 non-null  object 
 7   BONDAD_EMAIL      367705 non-null  int64  
 8   USU_TELF          367639 non-null  object 
 9   IPCASOS           367705 non-null  int64  
 10  IP_Country        345941 non-null  object 
 11  IP_Area           345941 non-null  object 
 12  USU_TIPO          84114 non-null   object 
 13  USU_TAMANIO       84116 non-null   object 
 14  USU_CIIU          84116 non-null   object 
 15  USU_ESTADO        84116 non-null   object 
 16  USU_DEPARTAMENTO  90508 non-null   object 
 17  FEC_CLIENTE       2613 non-null    object 
 18  FEC_ALTA          3711 non-null    object 
dtypes: float64(2), int64(4), object(13)
memory usage: 53.3+ MB


usuariosDF.head()


usuariosDF[['FEC_REGISTRO',
            'FEC_CLIENTE',
            'FEC_ALTA']] = usuariosDF[['FEC_REGISTRO', 
                                       'FEC_CLIENTE', 
                                       'FEC_ALTA']].apply(pd.to_datetime)

usuariosDF[['IDUSUARIO']] = usuariosDF[['IDUSUARIO']].astype(int).astype(str)


consumosDF = pd.read_csv('/Users/inigoab/Documents/DS_UOC/TFM/Potenciales_Compradores_ecommerce/Linea1-Consumos_Promocionales.txt',
    sep=";", decimal=",", encoding = "ISO-8859-1")

#consumosDF = pd.read_csv("./drive/MyDrive/Potenciales Compradores ecommerce/Linea1-Consumos Promocionales.txt", 
#                         sep=";", decimal=",", encoding = "ISO-8859-1")


consumosDF.drop(columns=['IDGRUPOPROD', 'DESCGRUPOPROD'], inplace=True)


consumosDF.head()


consumosDF[['IDPRODUCTO']] = consumosDF[['IDPRODUCTO']].astype(int).astype(str)
consumosDF[['IDCONSUMO']] = consumosDF[['IDCONSUMO']].astype(int).astype(str)
consumosDF[['IDUSUARIO']] = consumosDF[['IDUSUARIO']].astype(int).astype(str)

consumosDF['FECHACONSUMO'] =  pd.to_datetime(consumosDF['FECHACONSUMO'])


consumosDF.insert(2, 'FEC_CLIENTE', consumosDF['IDUSUARIO'].map(usuariosDF.set_index('IDUSUARIO')['FEC_CLIENTE']))
consumosDF[consumosDF['FEC_CLIENTE'].notnull()].head()


consumosDF[consumosDF['FEC_CLIENTE'] < consumosDF['FECHACONSUMO']]


consumosDF.drop(consumosDF[consumosDF['FEC_CLIENTE'] < consumosDF['FECHACONSUMO']].index, inplace=True)


len(consumosDF[consumosDF['FEC_CLIENTE'] < consumosDF['FECHACONSUMO']])

0


consumos_fechas = pd.concat([consumosDF.groupby(['IDUSUARIO'])['FECHACONSUMO'].min(), 
                             consumosDF.groupby(['IDUSUARIO'])['FECHACONSUMO'].max()], 
                            axis=1, keys=('primer_consumo','último_consumo')).reset_index()

consumos_fechas


consumosDF.groupby(['IDUSUARIO'])['DESCPRODUCTO'].value_counts().unstack(fill_value=0)\
    .sort_values(by='Ficha Básica')


consumosDF.groupby(['IDUSUARIO','DESCPRODUCTO']).size().unstack(fill_value=0)\
    .sort_values(by='Perfil Promocional')


consumosDF.groupby(['IDUSUARIO','DESCPRODUCTO'])['EMPCONSUL_EST'].value_counts()\
  .unstack('DESCPRODUCTO',fill_value=0)


consumosDF['DESCPRODUCTO'].value_counts()

Perfil Promocional    424032
Ficha Básica          371988
Name: DESCPRODUCTO, dtype: int64


consumosDF['EMPCONSUL_EST'].value_counts()

ACTIVA                                              656798
CANCELACION                                          81575
LIQUIDACION                                          35595
LEY DE INSOLVENCIA (REORGANIZACION EMPRESARIAL)       8475
EXTINGUIDA                                            6379
INACTIVA TEMPORAL                                     3079
REESTRUCTURACION O CONCORDATO                          434
INTERVENIDA                                            430
COINCIDENCIA HOMOGRAFA LISTA CLINTON (SDNT OFAC)        98
ABSORBIDA                                               19
ANULACIËN LIQUIDACIËN                                   17
SALIDA CLINTON (SDNT OFAC)                               7
DISUELTA                                                 3
Name: EMPCONSUL_EST, dtype: int64


consumosDF.info()

<class 'pandas.core.frame.DataFrame'>
Int64Index: 796020 entries, 0 to 796736
Data columns (total 10 columns):
 #   Column          Non-Null Count   Dtype         
---  ------          --------------   -----         
 0   IDCONSUMO       796020 non-null  object        
 1   IDUSUARIO       796020 non-null  object        
 2   FEC_CLIENTE     12752 non-null   datetime64[ns]
 3   IDPRODUCTO      796020 non-null  object        
 4   DESCPRODUCTO    796020 non-null  object        
 5   FECHACONSUMO    796020 non-null  datetime64[ns]
 6   EMPCONSUL_ID    796020 non-null  int64         
 7   EMPCONSUL_CIIU  792909 non-null  object        
 8   EMPCONSUL_PROV  792909 non-null  object        
 9   EMPCONSUL_EST   792909 non-null  object        
dtypes: datetime64[ns](2), int64(1), object(7)
memory usage: 66.8+ MB


consumosDF.isnull().sum()

IDCONSUMO              0
IDUSUARIO              0
FEC_CLIENTE       783268
IDPRODUCTO             0
DESCPRODUCTO           0
FECHACONSUMO           0
EMPCONSUL_ID           0
EMPCONSUL_CIIU      3111
EMPCONSUL_PROV      3111
EMPCONSUL_EST       3111
dtype: int64


consumos_agrupados = consumosDF.groupby(['IDUSUARIO',
                                         'DESCPRODUCTO']).size().unstack(fill_value=0).reset_index()
consumos_agrupados


DF = usuariosDF.merge(consumos_agrupados, on='IDUSUARIO', how='left')\
    .merge(consumos_fechas, on='IDUSUARIO', how='left')
DF.head()


DF[['Ficha Básica','Perfil Promocional']].isnull().sum()

Ficha Básica          67
Perfil Promocional    67
dtype: int64


DF['Ficha Básica'].fillna(0, inplace=True)
DF['Perfil Promocional'].fillna(0, inplace=True)


DF['Ficha Básica'] = DF['Ficha Básica'].astype(int)
DF['Perfil Promocional'] = DF['Perfil Promocional'].astype(int)


loginsDF = pd.read_csv('/Users/inigoab/Documents/DS_UOC/TFM/Potenciales_Compradores_ecommerce/Linea1-Logins.txt',
    sep=";", decimal=",", encoding = "ISO-8859-1")

#loginsDF = pd.read_csv("./drive/MyDrive/Potenciales Compradores ecommerce/Linea1-Logins.txt", 
#                         sep=";", decimal=",", encoding = "ISO-8859-1")


loginsDF.head()


loginsDF[['IDUSUARIO']] = loginsDF[['IDUSUARIO']].astype(int).astype(str)

loginsDF['FECHALOGIN'] =  pd.to_datetime(loginsDF['FECHALOGIN'])


loginsDF.insert(2, 'FEC_CLIENTE', loginsDF['IDUSUARIO'].map(usuariosDF.set_index('IDUSUARIO')['FEC_CLIENTE']))
loginsDF[loginsDF['FEC_CLIENTE'].notnull()].head()


loginsDF[loginsDF['FEC_CLIENTE'] < loginsDF['FECHALOGIN']]


loginsDF.drop(loginsDF[loginsDF['FEC_CLIENTE'] < loginsDF['FECHALOGIN']].index, inplace=True)


loginsDF.info()

<class 'pandas.core.frame.DataFrame'>
Int64Index: 616894 entries, 1 to 617748
Data columns (total 3 columns):
 #   Column       Non-Null Count   Dtype         
---  ------       --------------   -----         
 0   IDUSUARIO    616894 non-null  object        
 1   FECHALOGIN   616894 non-null  datetime64[ns]
 2   FEC_CLIENTE  15859 non-null   datetime64[ns]
dtypes: datetime64[ns](2), object(1)
memory usage: 18.8+ MB


loginsDF['IDUSUARIO'].nunique()

197129


loginsDF.isnull().sum()

IDUSUARIO           0
FECHALOGIN          0
FEC_CLIENTE    601035
dtype: int64


logins_fechas = pd.concat([loginsDF.groupby(['IDUSUARIO'])['FECHALOGIN'].min(), 
                             loginsDF.groupby(['IDUSUARIO'])['FECHALOGIN'].max()], 
                            axis=1, keys=('primer_login','último_login')).reset_index()

logins_fechas


logins_agrupados = loginsDF.groupby(
                    ['IDUSUARIO']).size().reset_index(name='N_logins')

logins_agrupados


DF = DF.merge(logins_agrupados, on='IDUSUARIO', how='left')\
    .merge(logins_fechas, on='IDUSUARIO', how='left')

DF


DF[['N_logins']].isnull().sum()

N_logins    170625
dtype: int64


DF['N_logins'].fillna(0, inplace=True)


DF['N_logins'] = DF['N_logins'].astype(int)


sesionesDF = pd.read_csv('/Users/inigoab/Documents/DS_UOC/TFM/Potenciales_Compradores_ecommerce/Linea1-Sesiones.txt',
    sep=";", decimal=",", encoding = "ISO-8859-1")

#sesionesDF = pd.read_csv("./drive/MyDrive/Potenciales Compradores ecommerce/Linea1-Sesiones.txt", 
#                         sep=";", decimal=",", encoding = "ISO-8859-1")


sesionesDF.head()


sesionesDF[['IDUSUARIO']] = sesionesDF[['IDUSUARIO']].astype(int).astype(str)
sesionesDF['FECHA_SESION'] =  pd.to_datetime(sesionesDF['FECHA_SESION'])


sesionesDF.insert(3, 'FEC_CLIENTE', sesionesDF['IDUSUARIO'].map(usuariosDF.set_index('IDUSUARIO')['FEC_CLIENTE']))
sesionesDF[sesionesDF['FEC_CLIENTE'].notnull()].head()


sesionesDF[sesionesDF['FEC_CLIENTE'] < sesionesDF['FECHA_SESION']]


sesionesDF.drop(sesionesDF[sesionesDF['FEC_CLIENTE'] < sesionesDF['FECHA_SESION']].index, inplace=True)


sesionesDF.info()

<class 'pandas.core.frame.DataFrame'>
Int64Index: 407757 entries, 0 to 408027
Data columns (total 4 columns):
 #   Column        Non-Null Count   Dtype         
---  ------        --------------   -----         
 0   IDUSUARIO     407757 non-null  object        
 1   FECHA_SESION  407757 non-null  datetime64[ns]
 2   SESIONES      407757 non-null  int64         
 3   FEC_CLIENTE   4165 non-null    datetime64[ns]
dtypes: datetime64[ns](2), int64(1), object(1)
memory usage: 15.6+ MB


sesionesDF['IDUSUARIO'].nunique()

367702


sesionesDF.isnull().sum()

IDUSUARIO            0
FECHA_SESION         0
SESIONES             0
FEC_CLIENTE     403592
dtype: int64


sesiones_fechas = pd.concat([sesionesDF.groupby(['IDUSUARIO'])['FECHA_SESION'].min(), 
                             sesionesDF.groupby(['IDUSUARIO'])['FECHA_SESION'].max()], 
                            axis=1, keys=('primera_sesión','última_sesión')).reset_index()

sesiones_fechas


sesiones_agrupadas = sesionesDF.groupby(
    ['IDUSUARIO'])['SESIONES'].sum().reset_index(name='N_sesiones')
sesiones_agrupadas.head()


DF = DF.merge(sesiones_agrupadas, on='IDUSUARIO', how='left')\
    .merge(sesiones_fechas, on='IDUSUARIO', how='left')

DF


DF[['N_sesiones']].isnull().sum()

N_sesiones    3
dtype: int64


DF['N_sesiones'].fillna(0, inplace=True)


DF['N_sesiones'] = DF['N_sesiones'].astype(int)


## VENTAS


'''
ventasDF = pd.read_csv(
    '/Users/inigoab/Documents/DS_UOC/TFM/Potenciales_Compradores_ecommerce/Linea1-Ventas.txt',
    sep=";", decimal=",", encoding = "ISO-8859-1")

ventasDF[['FECHAVENTA']] = ventasDF['FECHAVENTA'].apply(pd.to_datetime)
ventasDF[['IDVENTA','ID_USUARIO']] = ventasDF[['IDVENTA','ID_USUARIO']].astype(int).astype(str)
ventasDF[['IMPORTES']] = ventasDF[['IMPORTES']].astype(int)
ventasDF.loc[:,('VP Informe',
                'BONO', 
                'SUSCRIPCION', 
                'VP Listado')] = ventasDF.loc[:,('VP Informe',
                                                 'BONO', 
                                                 'SUSCRIPCION', 
                                                 'VP Listado')].fillna(0)

ventasDF.head()

'''

'\nventasDF = pd.read_csv(\n    \'/Users/inigoab/Documents/DS_UOC/TFM/Potenciales_Compradores_ecommerce/Linea1-Ventas.txt\',\n    sep=";", decimal=",", encoding = "ISO-8859-1")\n\nventasDF[[\'FECHAVENTA\']] = ventasDF[\'FECHAVENTA\'].apply(pd.to_datetime)\nventasDF[[\'IDVENTA\',\'ID_USUARIO\']] = ventasDF[[\'IDVENTA\',\'ID_USUARIO\']].astype(int).astype(str)\nventasDF[[\'IMPORTES\']] = ventasDF[[\'IMPORTES\']].astype(int)\nventasDF.loc[:,(\'VP Informe\',\n                \'BONO\', \n                \'SUSCRIPCION\', \n                \'VP Listado\')] = ventasDF.loc[:,(\'VP Informe\',\n                                                 \'BONO\', \n                                                 \'SUSCRIPCION\', \n                                                 \'VP Listado\')].fillna(0)\n\nventasDF.head()\n\n'


counter = Counter(DF['IND_CLIENTE'].values)

for k,v in counter.items():
    per = v / sum(counter.values()) * 100
    print('Class= %d, Count= %d, Percentage= %.2f%%' % (k, v, per))

    
ax = sns.countplot(x="IND_CLIENTE", data=DF)
plt.title('Porcentaje de usuarios vs clientes', fontsize=20)
total = len(usuariosDF['IND_CLIENTE'])
for p in ax.patches:
    percentage = '{:.2f}%'.format(100 * p.get_height()/total)
    x = p.get_x() + p.get_width() / 2 - 0.05
    y = p.get_y() + p.get_height()
    ax.annotate(percentage, (x, y), size = 14)

plt.show()

Class= 0, Count= 365086, Percentage= 99.29%
Class= 1, Count= 2619, Percentage= 0.71%


# Tomo esta tasa de conversión como la de referencia
r_conversion = list(counter.values())[1]/ sum(counter.values())


plt.pie([float(v) for v in counter.values()], labels=[float(k) for k in counter],
           autopct='%1.2f%%')
plt.title('Porcentaje de usuarios vs clientes', fontsize=20)
plt.show()


DF.select_dtypes('datetime').max()

FEC_REGISTRO     2019-12-31
FEC_CLIENTE      2021-04-01
FEC_ALTA         2021-11-01
primer_consumo   2021-05-01
último_consumo   2021-12-01
primer_login     2021-12-06
último_login     2021-12-06
primera_sesión   2019-12-31
última_sesión    2022-12-01
dtype: datetime64[ns]


print('{} usuarios se registraron con fecha previa a ser clientes,\
 {} con fecha posterior y {} en la misma fecha.'.format(
    len(DF[usuariosDF['FEC_REGISTRO'] < DF['FEC_CLIENTE']]),
    len(DF[usuariosDF['FEC_REGISTRO'] > DF['FEC_CLIENTE']]),
    len(DF[usuariosDF['FEC_REGISTRO'] == DF['FEC_CLIENTE']])))

1057 usuarios se registraron con fecha previa a ser clientes, 161 con fecha posterior y 1395 en la misma fecha.


DF[['FEC_REGISTRO', 'FEC_CLIENTE']].sort_values(by='FEC_CLIENTE', ascending=False)


DF['FEC_REGISTRO'].value_counts().to_frame('registros').resample('W').mean()\
    .plot(figsize=(18, 4), title='Nº de registros por fecha')
DF['FEC_CLIENTE'].value_counts().to_frame('clientes').resample('W').mean()\
    .plot(figsize=(18, 4), title='Nº de nuevos clientes por fecha')

plt.show()


DF['FEC_REGISTRO'].drop(DF[DF['FEC_CLIENTE'] <= DF['FEC_REGISTRO']].index)\
    .value_counts().to_frame('registros').resample('W').mean().plot(figsize=(18, 4), 
                                                                    title='Nº de registros por fecha')
DF['FEC_CLIENTE'].drop(DF[DF['FEC_CLIENTE'] <= DF['FEC_REGISTRO']].index)\
    .value_counts().to_frame('clientes').resample('W').mean().plot(figsize=(18, 4), 
                                                                   title='Nº de nuevos clientes por fecha')

plt.show()


pd.DataFrame(DF[(DF['IND_CLIENTE'] == 1) &
               (DF['FEC_CLIENTE'] > DF['FEC_REGISTRO'])]['FEC_CLIENTE'] 
             - 
             DF[(DF['IND_CLIENTE'] == 1) & (DF['FEC_CLIENTE'] > DF['FEC_REGISTRO'])]['FEC_REGISTRO'],
             columns=['días']).sort_values(by='días').value_counts().to_frame().hist(bins=20, figsize=(18,5))

plt.show()


pd.DataFrame(DF[(DF['IND_CLIENTE'] == 1) &
               (DF['FEC_CLIENTE'] > DF['FEC_REGISTRO'])]['FEC_CLIENTE'] 
             - 
             DF[(DF['IND_CLIENTE'] == 1) & (DF['FEC_CLIENTE'] > DF['FEC_REGISTRO'])]['FEC_REGISTRO'],
             columns=['días']).sort_values(by='días')[['días']].astype('int').boxplot(figsize=(18,5),vert=False)

plt.show()

/Users/inigoab/opt/anaconda3/envs/TFM/lib/python3.7/site-packages/ipykernel_launcher.py:5: FutureWarning: casting timedelta64[ns] values to int64 with .astype(...) is deprecated and will raise in a future version. Use .view(...) instead.
  """


pd.DataFrame(DF[(DF['IND_CLIENTE'] == 1) &
               (DF['FEC_CLIENTE'] > DF['FEC_REGISTRO'])]['FEC_CLIENTE'] 
             - 
             DF[(DF['IND_CLIENTE'] == 1) & (DF['FEC_CLIENTE'] > DF['FEC_REGISTRO'])]['FEC_REGISTRO'],
             columns=['días']).sort_values(by='días').describe()


pd.DataFrame(DF[(DF['IND_CLIENTE'] == 1)]['FEC_CLIENTE'] 
             - 
             DF[(DF['IND_CLIENTE'] == 1)]['FEC_REGISTRO'],
             columns=['días']).dropna().sort_values(by='días', ascending=False)[['días']]\
            .astype('int').boxplot(figsize=(18,5),vert=False)

plt.show()

/Users/inigoab/opt/anaconda3/envs/TFM/lib/python3.7/site-packages/ipykernel_launcher.py:5: FutureWarning: casting timedelta64[ns] values to int64 with .astype(...) is deprecated and will raise in a future version. Use .view(...) instead.
  """


pd.DataFrame(DF[(DF['IND_CLIENTE'] == 1)]['FEC_CLIENTE'] 
             - 
             DF[(DF['IND_CLIENTE'] == 1)]['FEC_REGISTRO'],
             columns=['días']).dropna().sort_values(by='días', ascending=False)\
            .describe()


DF.select_dtypes('datetime').columns

Index(['FEC_REGISTRO', 'FEC_CLIENTE', 'FEC_ALTA', 'primer_consumo',
       'último_consumo', 'primer_login', 'último_login', 'primera_sesión',
       'última_sesión'],
      dtype='object')


print((DF['último_consumo'] - DF['primer_consumo']).isnull().sum())
print((DF['último_login'] - DF['primer_login']).isnull().sum())
print((DF['última_sesión'] - DF['primera_sesión']).isnull().sum())

67
170625
3


print((DF['primer_login']).isnull().sum())
print((DF['último_login']).isnull().sum())

170625
170625


DF[DF['primer_login'].isnull()].groupby('IND_CLIENTE')['IDUSUARIO'].count()

IND_CLIENTE
0    170437
1       188
Name: IDUSUARIO, dtype: int64


DF['díasEntreConsumos'] = (DF['último_consumo'] - DF['primer_consumo'])\
                            .dt.days.fillna(0).astype(int)

DF['díasEntreLogins'] = (DF['último_login'] - DF['primer_login'])\
                            .dt.days.fillna(0).astype(int)

DF['díasEntreSesiones'] = (DF['última_sesión'] - DF['primera_sesión'])\
                            .dt.days.fillna(0).astype(int)


DF.info()

<class 'pandas.core.frame.DataFrame'>
Int64Index: 367705 entries, 0 to 367704
Data columns (total 32 columns):
 #   Column              Non-Null Count   Dtype         
---  ------              --------------   -----         
 0   IDUSUARIO           367705 non-null  object        
 1   TIPOUSUARIO         367705 non-null  object        
 2   FEC_REGISTRO        367705 non-null  datetime64[ns]
 3   CANAL_REGISTRO      360171 non-null  float64       
 4   IND_CLIENTE         367705 non-null  int64         
 5   IND_ALTA            367705 non-null  int64         
 6   TIPOEMAIL           367705 non-null  object        
 7   BONDAD_EMAIL        367705 non-null  int64         
 8   USU_TELF            367639 non-null  object        
 9   IPCASOS             367705 non-null  int64         
 10  IP_Country          345941 non-null  object        
 11  IP_Area             345941 non-null  object        
 12  USU_TIPO            84114 non-null   object        
 13  USU_TAMANIO         84116 non-null   object        
 14  USU_CIIU            84116 non-null   object        
 15  USU_ESTADO          84116 non-null   object        
 16  USU_DEPARTAMENTO    90508 non-null   object        
 17  FEC_CLIENTE         2613 non-null    datetime64[ns]
 18  FEC_ALTA            3711 non-null    datetime64[ns]
 19  Ficha Básica        367705 non-null  int64         
 20  Perfil Promocional  367705 non-null  int64         
 21  primer_consumo      367638 non-null  datetime64[ns]
 22  último_consumo      367638 non-null  datetime64[ns]
 23  N_logins            367705 non-null  int64         
 24  primer_login        197080 non-null  datetime64[ns]
 25  último_login        197080 non-null  datetime64[ns]
 26  N_sesiones          367705 non-null  int64         
 27  primera_sesión      367702 non-null  datetime64[ns]
 28  última_sesión       367702 non-null  datetime64[ns]
 29  díasEntreConsumos   367705 non-null  int64         
 30  díasEntreLogins     367705 non-null  int64         
 31  díasEntreSesiones   367705 non-null  int64         
dtypes: datetime64[ns](9), float64(1), int64(11), object(11)
memory usage: 100.6+ MB


pd.concat([DF.isnull().sum().to_frame('Nº nulos'),
          (DF.isnull().sum()/DF.shape[0]).to_frame('% nulos').mul(100).round(2).astype(str) + '%'],
          axis=1)


sns.displot(
    data=DF.isna().melt(value_name="missing"),
    y="variable",
    hue="missing",
    multiple="fill"
)
plt.show()


# Elimino los campos con más de un 30% de valores nulos. En este caso se trata de los de tipo USU porque tienen más de un 75% de valores nulos.
DF.dropna(thresh=0.7*len(DF), axis='columns', inplace = True)


# Elimino IDUSUARIO porque sólo servía para juntar las tablas
DF.drop(columns=['IDUSUARIO'], inplace=True)

# Elimino las fechas porque ya las he usado para crear nuevas variables
DF.drop(columns=DF.select_dtypes('datetime').columns, inplace=True)

# Elimino USU_TELF ya que no parece útil debido a las portabilidades
DF.drop(columns=['USU_TELF'], inplace=True)

# Las direcciones de IP no son fiables porque es muy fácil que estén enmascaradas
DF.drop(columns=['IP_Country','IP_Area'], inplace=True)

# Elimino el indicador de alta ya que se indica en el enunciado que está muy correlacionado con el indicador de cliente
DF.drop(columns=['IND_ALTA'], inplace=True)


DF.info()

<class 'pandas.core.frame.DataFrame'>
Int64Index: 367705 entries, 0 to 367704
Data columns (total 13 columns):
 #   Column              Non-Null Count   Dtype  
---  ------              --------------   -----  
 0   TIPOUSUARIO         367705 non-null  object 
 1   CANAL_REGISTRO      360171 non-null  float64
 2   IND_CLIENTE         367705 non-null  int64  
 3   TIPOEMAIL           367705 non-null  object 
 4   BONDAD_EMAIL        367705 non-null  int64  
 5   IPCASOS             367705 non-null  int64  
 6   Ficha Básica        367705 non-null  int64  
 7   Perfil Promocional  367705 non-null  int64  
 8   N_logins            367705 non-null  int64  
 9   N_sesiones          367705 non-null  int64  
 10  díasEntreConsumos   367705 non-null  int64  
 11  díasEntreLogins     367705 non-null  int64  
 12  díasEntreSesiones   367705 non-null  int64  
dtypes: float64(1), int64(10), object(2)
memory usage: 47.3+ MB


#prof = ProfileReport(DF)
#prof.to_file(output_file='report_DF.html')


pd.concat([DF["TIPOUSUARIO"].value_counts(),
           DF["TIPOUSUARIO"].value_counts(normalize=True).mul(100).round(2).astype(str) + '%'],
          axis=1, keys=('counts','percentage'))


pd.concat([DF.groupby(["IND_CLIENTE"])
           ["TIPOUSUARIO"].value_counts(),
           DF.groupby(["IND_CLIENTE"])
           ["TIPOUSUARIO"].value_counts(normalize=True).mul(100).round(2).astype(str) + '%'],
          axis=1, keys=('counts','%'))


DF.groupby(["TIPOUSUARIO", "IND_CLIENTE"]).size().unstack()\
        .plot.pie(autopct='%1.2f%%', subplots=True, legend=False, figsize=(11, 6))
plt.show()


pd.concat([DF.groupby(['TIPOUSUARIO'])
           ['IND_CLIENTE'].value_counts(),
           DF.groupby(['TIPOUSUARIO'])
           ['IND_CLIENTE'].value_counts(normalize=True).mul(100).round(2).astype(str) + '%'],
          axis=1, keys=('counts','%'))


DF.groupby(['IND_CLIENTE', 'TIPOUSUARIO']).size().unstack()\
        .plot.pie(autopct='%1.2f%%', subplots=True, legend=False, figsize=(18, 10))
plt.show()


sorted(DF['CANAL_REGISTRO'].unique())

[1.0, 2.0, 3.0, 4.0, 6.0, 7.0, 8.0, 9.0, nan]


DF['CANAL_REGISTRO'].isnull().sum()

7534


# Inputo los valores nulos (nan) como "otros_canales"

DF['CANAL_REGISTRO'].fillna("otros_canales", inplace=True)


DF['CANAL_REGISTRO'].loc[DF['CANAL_REGISTRO'] == 1.] = "SEM"
DF['CANAL_REGISTRO'].loc[DF['CANAL_REGISTRO'].isin([2.,3.,7.])] = "canales_populares"
DF['CANAL_REGISTRO'].loc[DF['CANAL_REGISTRO'].isin([6.,8.,9.])] = "canales_especializados"
DF['CANAL_REGISTRO'].loc[DF['CANAL_REGISTRO'].isin([6.,8.,9.])] = "canales_especializados"
DF['CANAL_REGISTRO'].loc[DF['CANAL_REGISTRO'] == 4.] = "SE0"

/Users/inigoab/opt/anaconda3/envs/TFM/lib/python3.7/site-packages/pandas/core/indexing.py:1732: SettingWithCopyWarning: 
A value is trying to be set on a copy of a slice from a DataFrame

See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy
  self._setitem_single_block(indexer, value, name)


pd.concat([DF['CANAL_REGISTRO'].value_counts(),
          DF['CANAL_REGISTRO'].value_counts(normalize=True).mul(100).round(2).astype(str) + '%'],
          axis=1, keys=('counts','%'))


DF['TIPOEMAIL'].isnull().sum()

0


# distribución de los distintos tipos de emails
pd.concat([DF['TIPOEMAIL'].value_counts(),
           DF['TIPOEMAIL'].value_counts(normalize=True).mul(100).round(2).astype(str) + '%'],
          axis=1, keys=('counts','percentage')).sort_values(by=['counts'], ascending=False).head(15)


# distribución de los distintos tipos de email entre los clientes
pd.concat([DF[DF['IND_CLIENTE']==1]['TIPOEMAIL'].value_counts(),
           DF[DF['IND_CLIENTE']==1]['TIPOEMAIL'].value_counts(normalize=True).mul(100).round(2).astype(str) + '%'],
          axis=1, keys=('counts','percentage')).sort_values(by=['counts'], ascending=False).head(15)


len(DF['TIPOEMAIL'].loc[DF['TIPOEMAIL'].str.contains('(^|\.)edu(\.|$)')])

/Users/inigoab/opt/anaconda3/envs/TFM/lib/python3.7/site-packages/ipykernel_launcher.py:1: UserWarning: This pattern has match groups. To actually get the groups, use str.extract.
  """Entry point for launching an IPython kernel.

11058


DF['TIPOEMAIL'].loc[DF['TIPOEMAIL'].str.contains('(^|\.)edu(\.|$)')] = 'email_educativo'

/Users/inigoab/opt/anaconda3/envs/TFM/lib/python3.7/site-packages/ipykernel_launcher.py:1: UserWarning: This pattern has match groups. To actually get the groups, use str.extract.
  """Entry point for launching an IPython kernel.
/Users/inigoab/opt/anaconda3/envs/TFM/lib/python3.7/site-packages/pandas/core/indexing.py:1732: SettingWithCopyWarning: 
A value is trying to be set on a copy of a slice from a DataFrame

See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy
  self._setitem_single_block(indexer, value, name)


len(DF['TIPOEMAIL'].loc[DF['TIPOEMAIL'].str.contains('(^|\.)co(\.|$)')])

/Users/inigoab/opt/anaconda3/envs/TFM/lib/python3.7/site-packages/ipykernel_launcher.py:1: UserWarning: This pattern has match groups. To actually get the groups, use str.extract.
  """Entry point for launching an IPython kernel.

15290


DF['TIPOEMAIL'].loc[DF['TIPOEMAIL'].str.contains('(^|\.)co(\.|$)')] = 'email_colombiano'

/Users/inigoab/opt/anaconda3/envs/TFM/lib/python3.7/site-packages/ipykernel_launcher.py:1: UserWarning: This pattern has match groups. To actually get the groups, use str.extract.
  """Entry point for launching an IPython kernel.


len(DF['TIPOEMAIL'].loc[DF['TIPOEMAIL'].str.contains('(^|\.)es(\.|$)')])

/Users/inigoab/opt/anaconda3/envs/TFM/lib/python3.7/site-packages/ipykernel_launcher.py:1: UserWarning: This pattern has match groups. To actually get the groups, use str.extract.
  """Entry point for launching an IPython kernel.

11295


DF['TIPOEMAIL'].loc[DF['TIPOEMAIL'].str.contains('(^|\.)es(\.|$)')] = 'email_español'

/Users/inigoab/opt/anaconda3/envs/TFM/lib/python3.7/site-packages/ipykernel_launcher.py:1: UserWarning: This pattern has match groups. To actually get the groups, use str.extract.
  """Entry point for launching an IPython kernel.


DF['TIPOEMAIL'].loc[DF['TIPOEMAIL'].str.contains('(?!(^gmail\.com$))(^|\.)gmail\.|(^|\.)google\.')] = 'email_google'

/Users/inigoab/opt/anaconda3/envs/TFM/lib/python3.7/site-packages/ipykernel_launcher.py:1: UserWarning: This pattern has match groups. To actually get the groups, use str.extract.
  """Entry point for launching an IPython kernel.


DF['TIPOEMAIL'].loc[DF['TIPOEMAIL'].str.contains(
    '(?!(^hotmail\.com$))(^|\.)hotmail\.|(^|\.)outlook\.|(^|\.)microsoft\.|(^|\.)live\.')] = 'email_microsoft'

/Users/inigoab/opt/anaconda3/envs/TFM/lib/python3.7/site-packages/ipykernel_launcher.py:2: UserWarning: This pattern has match groups. To actually get the groups, use str.extract.


# distribución de los distintos tipos de emails
pd.concat([DF['TIPOEMAIL'].value_counts(),
           DF['TIPOEMAIL'].value_counts(normalize=True).mul(100).round(2).astype(str) + '%'],
          axis=1, keys=('counts','percentage')).sort_values(by=['counts'], ascending=False).head(15)


frecuencias_email = DF['TIPOEMAIL'].value_counts(normalize=True)


#Agrupo los emails muy minoritarios (los que tienen una frecuencia inferior al 0,001%)
DF['TIPOEMAIL'] = DF['TIPOEMAIL'].mask(
    DF['TIPOEMAIL'].map(frecuencias_email) < 0.00001, 'email_muy_minoritario')


#Agrupo los emails minoritarios (los que tienen una frecuencia inferior al 0,1%)
DF['TIPOEMAIL'] = DF['TIPOEMAIL'].mask(
    DF['TIPOEMAIL'].map(frecuencias_email) < 0.001, 'email_minoritario')


#Agrupo los emails que tienen una frecuencia inferior al 1%
DF['TIPOEMAIL'] = DF['TIPOEMAIL'].mask(
    DF['TIPOEMAIL'].map(frecuencias_email) < 0.01, 'otros_email')


# distribución de los distintos tipos de emails
pd.concat([DF['TIPOEMAIL'].value_counts(),
           DF['TIPOEMAIL'].value_counts(normalize=True).mul(100).round(2).astype(str) + '%'],
          axis=1, keys=('counts','percentage')).sort_values(by=['counts'], ascending=False).head(15)


# distribución de los distintos tipos de email entre los clientes
pd.concat([DF[DF['IND_CLIENTE']==1]['TIPOEMAIL'].value_counts(),
           DF[DF['IND_CLIENTE']==1]['TIPOEMAIL'].value_counts(normalize=True).mul(100).round(2).astype(str) + '%'],
          axis=1, keys=('counts','percentage')).sort_values(by=['counts'], ascending=False).head(15)


#g = sns.countplot(data=DF, x='TIPOEMAIL', hue='IND_CLIENTE',stat = 'density')
#g.set_yscale("log")
#_ = g.set(xlabel="Class", ylabel="Survived")

sns.histplot(data=DF, x='TIPOEMAIL', hue='IND_CLIENTE', multiple="dodge", 
                  stat = 'proportion', shrink = 0.8, common_norm=True, log_scale=(False, True))
plt.xticks(rotation=75)

([0, 1, 2, 3, 4, 5, 6, 7, 8, 9],
 [Text(0, 0, ''),
  Text(0, 0, ''),
  Text(0, 0, ''),
  Text(0, 0, ''),
  Text(0, 0, ''),
  Text(0, 0, ''),
  Text(0, 0, ''),
  Text(0, 0, ''),
  Text(0, 0, ''),
  Text(0, 0, '')])


pd.concat([DF.groupby(['TIPOEMAIL'])
           ['IND_CLIENTE'].value_counts(),
           DF.groupby(['TIPOEMAIL'])
           ['IND_CLIENTE'].value_counts(normalize=True).mul(100).round(2).astype(str) + '%'],
          axis=1, keys=('counts','percentage'))


pd.concat([DF.groupby(['BONDAD_EMAIL'])
           ['IND_CLIENTE'].value_counts(),
           DF.groupby(['BONDAD_EMAIL'])
           ['IND_CLIENTE'].value_counts(normalize=True).mul(100).round(2).astype(str) + '%'],
          axis=1, keys=('counts','percentage'))


DF['BONDAD_EMAIL'].loc[DF['BONDAD_EMAIL'] == 20] = "email_OK"
DF['BONDAD_EMAIL'].loc[DF['BONDAD_EMAIL'] == 9] = "email_Naranja"
DF['BONDAD_EMAIL'].loc[DF['BONDAD_EMAIL'] == 1] = "SPAM"
DF['BONDAD_EMAIL'].loc[DF['BONDAD_EMAIL'] == 0] = "email_Rojo"
DF['BONDAD_EMAIL'].loc[DF['BONDAD_EMAIL'] == -10] = "email_sin_dominio"
DF['BONDAD_EMAIL'].loc[DF['BONDAD_EMAIL'] == -20] = "No_email"

/Users/inigoab/opt/anaconda3/envs/TFM/lib/python3.7/site-packages/pandas/core/indexing.py:1732: SettingWithCopyWarning: 
A value is trying to be set on a copy of a slice from a DataFrame

See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy
  self._setitem_single_block(indexer, value, name)


DF['BONDAD_EMAIL'].unique()

array(['email_Rojo', 'email_OK', 'email_sin_dominio', 'SPAM', 'No_email',
       'email_Naranja'], dtype=object)


pd.concat([DF.groupby(['IND_CLIENTE'])
           ['BONDAD_EMAIL'].value_counts(),
           DF.groupby(['IND_CLIENTE'])
           ['BONDAD_EMAIL'].value_counts(normalize=True).mul(100).round(2).astype(str) + '%'],
          axis=1, keys=('counts','percentage'))


sns.histplot(data=DF, x='BONDAD_EMAIL', hue='IND_CLIENTE', multiple="dodge", 
                  stat = 'proportion', shrink = 0.8, common_norm=True, log_scale=(False, True))
plt.xticks(rotation=75)

([0, 1, 2, 3, 4, 5],
 [Text(0, 0, ''),
  Text(0, 0, ''),
  Text(0, 0, ''),
  Text(0, 0, ''),
  Text(0, 0, ''),
  Text(0, 0, '')])


DF.info()

<class 'pandas.core.frame.DataFrame'>
Int64Index: 367705 entries, 0 to 367704
Data columns (total 13 columns):
 #   Column              Non-Null Count   Dtype 
---  ------              --------------   ----- 
 0   TIPOUSUARIO         367705 non-null  object
 1   CANAL_REGISTRO      367705 non-null  object
 2   IND_CLIENTE         367705 non-null  int64 
 3   TIPOEMAIL           367705 non-null  object
 4   BONDAD_EMAIL        367705 non-null  object
 5   IPCASOS             367705 non-null  int64 
 6   Ficha Básica        367705 non-null  int64 
 7   Perfil Promocional  367705 non-null  int64 
 8   N_logins            367705 non-null  int64 
 9   N_sesiones          367705 non-null  int64 
 10  díasEntreConsumos   367705 non-null  int64 
 11  díasEntreLogins     367705 non-null  int64 
 12  díasEntreSesiones   367705 non-null  int64 
dtypes: int64(9), object(4)
memory usage: 47.3+ MB


DF.select_dtypes(object).info()

<class 'pandas.core.frame.DataFrame'>
Int64Index: 367705 entries, 0 to 367704
Data columns (total 4 columns):
 #   Column          Non-Null Count   Dtype 
---  ------          --------------   ----- 
 0   TIPOUSUARIO     367705 non-null  object
 1   CANAL_REGISTRO  367705 non-null  object
 2   TIPOEMAIL       367705 non-null  object
 3   BONDAD_EMAIL    367705 non-null  object
dtypes: object(4)
memory usage: 22.1+ MB


def test_chi2(dataset, variable_objetivo, alpha = 0.05):
    '''
    Dados el dataset y el nombre de la variable objetivo
    realiza el test chi2 sobre las variables categóricas.
    '''

    from scipy.stats import chi2_contingency

    print('- Hipótesis Nula (H0): No hay relación entre las variables')
    print('- Hipótesis Alternativa (H1): Hay relación entre las variables\n')

    X = dataset.select_dtypes(object)
    y = dataset[variable_objetivo]

    chi2 = []
    p_value = []
    chi2_check = []
    for i in X.columns:
        chi_res = chi2_contingency(pd.crosstab(y, X[i]))
        chi2.append(round(chi_res[0],4))
        p_value.append(chi_res[1])
        #print('p-valor', p_value[-1])
        if chi_res[1] < alpha:
            chi2_check.append('Se rechaza la hipótesis nula')
        else:
            chi2_check.append('No se rechaza la hipótesis nula')
      
    res = pd.DataFrame(data = [X.columns, chi2, p_value, chi2_check]).T 
      
    res.columns = ['Columna', 'Chi2','p-valor', 'Hipótesis']

    alpha
    quitar_columnas = res[res['p-valor'] > alpha]['Columna'].tolist()
    if len(quitar_columnas) > 0:
        for c in quitar_columnas:
            print('No se puede afirmar que haya relación entre {} y ser cliente, con un nivel de confianza del {}%.'. format(
                    c,round((1-alpha)*100)))
    else:
        print('No se puede afirmar que no exista relación entre las variables y ser cliente, con un nivel de confianza del {}%.'.format(
                round((1-alpha)*100)))
    print('\n')
  
    return res


test_chi2(DF, 'IND_CLIENTE')

- Hipótesis Nula (H0): No hay relación entre las variables
- Hipótesis Alternativa (H1): Hay relación entre las variables

No se puede afirmar que no exista relación entre las variables y ser cliente, con un nivel de confianza del 95%.


def corr_matrix(datos):
    '''
    Función para crear un gráfico de matriz de correlaciones entre las variables numéricas.
    '''
    corr = datos.drop(columns=datos.select_dtypes(object).columns).corr()

    sns.set_theme(style="white")
    mask = np.triu(np.ones_like(corr, dtype=bool))
    f, ax = plt.subplots(figsize=(11, 9))
    cmap = sns.diverging_palette(230, 20, as_cmap=True)

    sns.heatmap(corr, mask=mask, cmap=cmap, center=0,
                square=True, linewidths=.5, cbar_kws={"shrink": .5}, 
                annot=True, fmt='.2f',)
  
    plt.show()


corr_matrix(DF)


def test_anova(dataset, variable_objetivo, alpha = 0.05):
    
    from sklearn.feature_selection import SelectKBest
    from sklearn.feature_selection import f_classif
    
    print('- Hipótesis Nula (H0): No hay diferencia en la varizan en función de si el usuario es cliente o no')
    print('- Hipótesis Alternativa (H1): Hay diferencia en la varianza en función de si el usuario es cliente o no\n')

    X, y = dataset.select_dtypes(int).drop(columns=['IND_CLIENTE']), dataset[variable_objetivo]

    fs = SelectKBest(score_func=f_classif, k='all')
    fs.fit(X, y)
    
    anova_check = []
    for i in range(len(fs.scores_)):
        if fs.pvalues_[i] < alpha:
            anova_check.append('Se rechaza la hipótesis nula')
        else:
            anova_check.append('No se rechaza la hipótesis nula')
      
    res = pd.DataFrame(data = [X.select_dtypes(int).columns, fs.scores_, fs.pvalues_, anova_check]).T 
    res.columns = ['Columna', 'fvalor','p-valor', 'Hipótesis']

    quitar_columnas = res[res['p-valor'] > alpha]['Columna'].tolist()
    if len(quitar_columnas) > 0:
        for c in quitar_columnas:
            print('No se puede afirmar que haya diferencia en la varianza de {} y ser cliente, con un nivel de confianza del {}%.'. format(
                c,round((1-alpha)*100)))
    else:
        print('No se puede afirmar que no exista diferencia en la varianza de las variables y ser cliente, con un nivel de confianza del {}%.'.format(
            round((1-alpha)*100)))
    print('\n')
    
    return res


test_anova(DF, 'IND_CLIENTE')

- Hipótesis Nula (H0): No hay diferencia en la varizan en función de si el usuario es cliente o no
- Hipótesis Alternativa (H1): Hay diferencia en la varianza en función de si el usuario es cliente o no

No se puede afirmar que no exista diferencia en la varianza de las variables y ser cliente, con un nivel de confianza del 95%.


DF = DF.drop(columns=['Ficha Básica', 'díasEntreSesiones'])

DF


X = DF.drop(columns='IND_CLIENTE')
y = DF['IND_CLIENTE']

X_train, X_test1, y_train, y_test1 = train_test_split(X, y, test_size=0.30, random_state=42, stratify=y)

X = X_test1
y = y_test1

X_val, X_test, y_val, y_test = train_test_split(X, y, test_size=0.5, random_state=1, stratify=y)

#test = pd.concat([X_test, y_test], axis=1) # datos para test
#val = pd.concat([X_val, y_val], axis=1) # datos para validación
#train0 = pd.concat([X_train, y_train], axis=1) # en este no quito observaciones
#train = pd.concat([X_train, y_train], axis=1) # en este quito reduciré las observaciones

print('X_test', X_test.shape)
print('X_val', X_val.shape)
#print(train0.shape)
print('X_train', X_train.shape)

X_test (55156, 10)
X_val (55156, 10)
X_train (257393, 10)


def ohe(data):

    ohe = OneHotEncoder(sparse=False)
    data_ohencoded = pd.DataFrame(ohe.fit_transform(data.select_dtypes(object)))
    data_ohencoded.columns = np.unique(data.select_dtypes(object)).tolist()
  
    return pd.concat([data_ohencoded, data.reset_index(drop=True).drop(
        columns=data.select_dtypes(object).columns.tolist())], axis=1)


X_train = ohe(X_train)
X_val = ohe(X_val)
X_test = ohe(X_test)


X_train


X_train0 = X_train.copy()
X_val0 = X_val.copy()
X_test0 = X_test.copy()

#X_train0.to_pickle("./X_train0.pkl")  
#X_val0.to_pickle("./X_val0.pkl")  
#X_test0.to_pickle("./X_test0.pkl")


'''
X_train0 = pd.read_pickle("./X_train0.pkl")
X_val0 = pd.read_pickle("./X_val0.pkl")
X_test0 = pd.read_pickle("./X_test0.pkl")
'''

'\nX_train0 = pd.read_pickle("./X_train0.pkl")\nX_val0 = pd.read_pickle("./X_val0.pkl")\nX_test0 = pd.read_pickle("./X_test0.pkl")\n'


'''from sklearn.ensemble import RandomForestClassifier

#feature_names = [f"feature {i}" for i in range(X.shape[1])]
forest = RandomForestClassifier(random_state=0, )
forest.fit(X_train, y_train)

importances = forest.feature_importances_
std = np.std([tree.feature_importances_ for tree in forest.estimators_], axis=0)

forest_importances = pd.Series(importances, index=X_train.columns)

fig, ax = plt.subplots()
forest_importances.plot.bar(yerr=std, ax=ax)
ax.set_title("Feature importances using MDI")
ax.set_ylabel("Mean decrease in impurity")
fig.tight_layout()'''

'from sklearn.ensemble import RandomForestClassifier\n\n#feature_names = [f"feature {i}" for i in range(X.shape[1])]\nforest = RandomForestClassifier(random_state=0, )\nforest.fit(X_train, y_train)\n\nimportances = forest.feature_importances_\nstd = np.std([tree.feature_importances_ for tree in forest.estimators_], axis=0)\n\nforest_importances = pd.Series(importances, index=X_train.columns)\n\nfig, ax = plt.subplots()\nforest_importances.plot.bar(yerr=std, ax=ax)\nax.set_title("Feature importances using MDI")\nax.set_ylabel("Mean decrease in impurity")\nfig.tight_layout()'


'''
from sklearn.feature_selection import SelectFromModel

X.shape

clf = ExtraTreesClassifier(n_estimators=50)
clf = clf.fit(X, y)
clf.feature_importances_  

model = SelectFromModel(clf, prefit=True)
X_new = model.transform(X)
X_new.shape      '''

'\nfrom sklearn.feature_selection import SelectFromModel\n\nX.shape\n\nclf = ExtraTreesClassifier(n_estimators=50)\nclf = clf.fit(X, y)\nclf.feature_importances_  \n\nmodel = SelectFromModel(clf, prefit=True)\nX_new = model.transform(X)\nX_new.shape      '


'''%%time

from xgboost import XGBClassifier

counter = Counter(y_train)

xgb = XGBClassifier(
    scale_pos_weight= counter[0]/counter[1]
    )

#cv = RepeatedStratifiedKFold(n_splits=10, n_repeats=3, random_state=1)

rfecv = RFECV(estimator=xgb, 
              step=1, 
              cv=3,
              scoring='recall')

rfecv.fit(X_train, y_train)

cols_drop = X_train.columns[~rfecv.support_]
print('Se eliminan las columnas: ', cols_drop)'''

"%%time\n\nfrom xgboost import XGBClassifier\n\ncounter = Counter(y_train)\n\nxgb = XGBClassifier(\n    scale_pos_weight= counter[0]/counter[1]\n    )\n\n#cv = RepeatedStratifiedKFold(n_splits=10, n_repeats=3, random_state=1)\n\nrfecv = RFECV(estimator=xgb, \n              step=1, \n              cv=3,\n              scoring='recall')\n\nrfecv.fit(X_train, y_train)\n\ncols_drop = X_train.columns[~rfecv.support_]\nprint('Se eliminan las columnas: ', cols_drop)"


#X_train.columns[rfecv.support_]


%%time

rfe = RFECV(estimator=RandomForestClassifier(random_state=20, 
#                                             class_weight='balanced'
                                             ), step=1, cv=5, scoring='recall') 
rfe = rfe.fit(X_train0, y_train)

print('Se mantienen las columnas: \n', X_train0.columns[rfe.support_])

Se mantienen las columnas: 
 Index(['PF', 'SPAM', 'email_español', 'IPCASOS', 'Perfil Promocional',
       'N_logins', 'N_sesiones', 'díasEntreConsumos', 'díasEntreLogins'],
      dtype='object')
CPU times: user 1h 12min 33s, sys: 44.9 s, total: 1h 13min 18s
Wall time: 1h 16min 52s


%%time

rfe1 = RFECV(estimator=RandomForestClassifier(random_state=20, 
                                             class_weight='balanced'
                                             ), step=1, cv=5, scoring='recall') 
rfe1 = rfe.fit(X_train0, y_train)

print('Se mantienen las columnas: \n', X_train0.columns[rfe1.support_])

Se mantienen las columnas: 
 Index(['PF', 'SPAM', 'email_español', 'IPCASOS', 'Perfil Promocional',
       'N_logins', 'N_sesiones', 'díasEntreConsumos', 'díasEntreLogins'],
      dtype='object')
CPU times: user 1h 13min 29s, sys: 1min, total: 1h 14min 30s
Wall time: 1h 17min 33s


X_val = X_val0.drop(columns=X_train0.columns[~rfe.support_])
X_test = X_test0.drop(columns=X_train0.columns[~rfe.support_])
X_train = X_train0.drop(columns=X_train0.columns[~rfe.support_])


X_train0.shape

(257393, 30)


print(X_train.shape)
print(X_val.shape)
print(X_test.shape)

(257393, 9)
(55156, 9)
(55156, 9)


'''
X_train.to_pickle("./X_train.pkl")  
X_val.to_pickle("./X_val.pkl")  
X_test.to_pickle("./X_test.pkl")
'''

'\nX_train.to_pickle("./X_train.pkl")  \nX_val.to_pickle("./X_val.pkl")  \nX_test.to_pickle("./X_test.pkl")\n'


'''
X_train = pd.read_pickle("./X_train.pkl")
X_val = pd.read_pickle("./X_val.pkl")
X_test = pd.read_pickle("./X_test.pkl")
'''

'\nX_train = pd.read_pickle("./X_train.pkl")\nX_val = pd.read_pickle("./X_val.pkl")\nX_test = pd.read_pickle("./X_test.pkl")\n'


#unpickled_X_train = pd.read_pickle("./X_train.pkl") 
#unpickled_X_test = pd.read_pickle("./X_test.pkl")


#unpickled_X_train.equals(X_train)


#unpickled_X_test.equals(X_test)


dill.dump_session('TFM_notebook_env.db')


#import dill
#dill.load_session('TFM_notebook_env.db')


%%time

print(Counter(y_train))
X_ROS, y_ROS = RandomOverSampler(random_state=42).fit_resample(X_train, y_train)
print(Counter(y_ROS))
print('\n')
'''
X_ROS.to_pickle("./X_ROS.pkl")  
y_ROS.to_pickle("./y_ROS.pkl")
'''

Counter({0: 255560, 1: 1833})
Counter({0: 255560, 1: 255560})


CPU times: user 440 ms, sys: 57.1 ms, total: 497 ms
Wall time: 500 ms

'\nX_ROS.to_pickle("./X_ROS.pkl")  \ny_ROS.to_pickle("./y_ROS.pkl")\n'


%%time

print(Counter(y_train))
X_SMOTE, y_SMOTE = SMOTE(random_state=42).fit_resample(X_train, y_train)
print(Counter(y_SMOTE))
print('\n')
'''
X_SMOTE.to_pickle("./X_SMOTE.pkl")  
y_SMOTE.to_pickle("./y_SMOTE.pkl")
'''

Counter({0: 255560, 1: 1833})
Counter({0: 255560, 1: 255560})


CPU times: user 398 ms, sys: 82.6 ms, total: 481 ms
Wall time: 487 ms

'\nX_SMOTE.to_pickle("./X_SMOTE.pkl")  \ny_SMOTE.to_pickle("./y_SMOTE.pkl")\n'


%%time

from imblearn.over_sampling import SMOTENC
print(Counter(y_train))
X_SMOTENC, y_SMOTENC = SMOTENC(random_state=1,
                               categorical_features=[0, 1,2],
                              ).fit_resample(X_train, y_train)
print(Counter(y_SMOTENC))
print('\n')

Counter({0: 255560, 1: 1833})
Counter({0: 255560, 1: 255560})


CPU times: user 31.2 s, sys: 1.03 s, total: 32.2 s
Wall time: 32.1 s


%%time

print(Counter(y_train))
X_BS, y_BS = BorderlineSMOTE(random_state=42).fit_resample(X_train, y_train)
print(Counter(y_BS))
print('\n')
'''
X_BS.to_pickle("./X_BS.pkl")
y_BS.to_pickle("./y_BS.pkl") 
'''

Counter({0: 255560, 1: 1833})
Counter({0: 255560, 1: 255560})


CPU times: user 1min 24s, sys: 1.19 s, total: 1min 25s
Wall time: 1min 25s

'\nX_BS.to_pickle("./X_BS.pkl")\ny_BS.to_pickle("./y_BS.pkl") \n'


%%time

print(Counter(y_train))
X_SVMSMOTE, y_SVMSMOTE = SVMSMOTE(
    #sampling_strategy=0.3,
    random_state=42).fit_resample(X_train, y_train)
print(Counter(y_SVMSMOTE))
#print(Counter(y_SVMSMOTE)[1] / sum(Counter(y_SVMSMOTE).values()) * 100)
print('\n')

Counter({0: 255560, 1: 1833})
Counter({0: 255560, 1: 255560})


CPU times: user 6min 35s, sys: 2.21 s, total: 6min 37s
Wall time: 6min 42s


%%time

print(Counter(y_train))
X_ADASYN, y_ADASYN = ADASYN(random_state=42).fit_resample(X_train, y_train)
print(Counter(y_ADASYN))
print('\n')
'''
X_ADASYN.to_pickle("./X_ADASYN.pkl")  
y_ADASYN.to_pickle("./y_ADASYN.pkl")
'''

Counter({0: 255560, 1: 1833})
Counter({1: 255929, 0: 255560})


CPU times: user 1min 23s, sys: 1.06 s, total: 1min 24s
Wall time: 1min 24s

'\nX_ADASYN.to_pickle("./X_ADASYN.pkl")  \ny_ADASYN.to_pickle("./y_ADASYN.pkl")\n'


dill.dump_session('TFM_notebook_env.db')


%%time

print(Counter(y_train))
undersample_RandomUnderSampler = RandomUnderSampler(random_state=1)
X_RUS, y_RUS = undersample_RandomUnderSampler.fit_resample(X_train, y_train)
print(Counter(y_RUS))
print('\n')
'''
X_RUS.to_pickle("./X_RUS.pkl")  
y_RUS.to_pickle("./y_RUS.pkl") 
'''

Counter({0: 255560, 1: 1833})
Counter({0: 1833, 1: 1833})


CPU times: user 153 ms, sys: 18.3 ms, total: 171 ms
Wall time: 169 ms

'\nX_RUS.to_pickle("./X_RUS.pkl")  \ny_RUS.to_pickle("./y_RUS.pkl") \n'


%%time

print(Counter(y_train))
undersample_NearMiss1 = NearMiss(version=1, n_neighbors=3)
X_NM1, y_NM1 = undersample_NearMiss1.fit_resample(X_train, y_train)
print(Counter(y_NM1))
print('\n')
'''
X_NM1.to_pickle("./X_NM1.pkl")
y_NM1.to_pickle("./y_NM1.pkl")
'''

Counter({0: 255560, 1: 1833})
Counter({0: 1833, 1: 1833})


CPU times: user 3.83 s, sys: 64.9 ms, total: 3.9 s
Wall time: 3.9 s

'\nX_NM1.to_pickle("./X_NM1.pkl")\ny_NM1.to_pickle("./y_NM1.pkl")\n'


'''
# Consume demasiada RAM

%%time

print(Counter(y))
undersample_NearMiss2 = NearMiss(version=2, n_neighbors=3)
X_NM2, y_NM2 = undersample_NearMiss2.fit_resample(X_train, y_train)
print(Counter(y_NM2))
print('\n')

X_NM2.to_pickle("./X_NM2.pkl")  
y_NM2.to_pickle("./y_NM2.pkl")
'''

'\n# Consume demasiada RAM\n\n%%time\n\nprint(Counter(y))\nundersample_NearMiss2 = NearMiss(version=2, n_neighbors=3)\nX_NM2, y_NM2 = undersample_NearMiss2.fit_resample(X_train, y_train)\nprint(Counter(y_NM2))\nprint(\'\n\')\n\nX_NM2.to_pickle("./X_NM2.pkl")  \ny_NM2.to_pickle("./y_NM2.pkl")\n'


%%time

print(Counter(y_train))
undersample_NearMiss3 = NearMiss(version=3, n_neighbors_ver3=3)
X_NM3, y_NM3 = undersample_NearMiss3.fit_resample(X_train, y_train)
print(Counter(y_NM3))
print('\n')
'''
X_NM3.to_pickle("./X_NM3.pkl")  
y_NM3.to_pickle("./y_NM3.pkl")
'''

Counter({0: 255560, 1: 1833})
Counter({0: 1833, 1: 1833})


CPU times: user 1min 25s, sys: 1.04 s, total: 1min 26s
Wall time: 1min 26s

'\nX_NM3.to_pickle("./X_NM3.pkl")  \ny_NM3.to_pickle("./y_NM3.pkl")\n'


dill.dump_session('TFM_notebook_env.db')


%%time

from imblearn.under_sampling import ClusterCentroids
print(Counter(y_train))
X_CC, y_CC = ClusterCentroids(random_state=42).fit_resample(X_train, y_train)
print(Counter(y_CC))
print('\n')

Counter({0: 255560, 1: 1833})
Counter({0: 1833, 1: 1833})


CPU times: user 1h 14min 54s, sys: 4min 59s, total: 1h 19min 54s
Wall time: 28min 21s


'''
# demasiado tiempo de ejecución

print(Counter(y))
undersample_CondensedNearestNeighbour = CondensedNearestNeighbour(n_neighbors=1, random_state=42)
X_CNN, y_CNN = undersample_CondensedNearestNeighbour.fit_resample(X_train, y_train)
print(Counter(y_CNN))
print('\n')

X_CNN.to_pickle("./X_CNN.pkl")  
y_CNN.to_pickle("./y_CNN.pkl")
'''

'\n# demasiado tiempo de ejecución\n\nprint(Counter(y))\nundersample_CondensedNearestNeighbour = CondensedNearestNeighbour(n_neighbors=1, random_state=42)\nX_CNN, y_CNN = undersample_CondensedNearestNeighbour.fit_resample(X_train, y_train)\nprint(Counter(y_CNN))\nprint(\'\n\')\n\nX_CNN.to_pickle("./X_CNN.pkl")  \ny_CNN.to_pickle("./y_CNN.pkl")\n'


%%time

print(Counter(y_train))
undersample_TL = TomekLinks()
X_TL, y_TL = undersample_TL.fit_resample(X_train, y_train)
print(Counter(y_TL))
'''
X_TL.to_pickle("./X_TL.pkl")  
y_TL.to_pickle("./y_TL.pkl") 
'''

Counter({0: 255560, 1: 1833})
Counter({0: 255202, 1: 1833})
CPU times: user 3min 25s, sys: 1.56 s, total: 3min 26s
Wall time: 3min 28s

'\nX_TL.to_pickle("./X_TL.pkl")  \ny_TL.to_pickle("./y_TL.pkl") \n'


%%time

print(Counter(y_train))
undersample_ENN = EditedNearestNeighbours(n_neighbors=3)
X_ENN, y_ENN = undersample_ENN.fit_resample(X_train, y_train)
print(Counter(y_ENN))
'''
X_ENN.to_pickle("./X_ENN.pkl")  
y_ENN.to_pickle("./y_ENN.pkl")
'''

Counter({0: 255560, 1: 1833})
Counter({0: 253023, 1: 1833})
CPU times: user 3min 28s, sys: 1.49 s, total: 3min 29s
Wall time: 3min 30s

'\nX_ENN.to_pickle("./X_ENN.pkl")  \ny_ENN.to_pickle("./y_ENN.pkl")\n'


%%time

from imblearn.under_sampling import RepeatedEditedNearestNeighbours

print(Counter(y_train))
X_RENN, y_RENN = RepeatedEditedNearestNeighbours().fit_resample(X_train, y_train)
print(Counter(y_RENN))

Counter({0: 255560, 1: 1833})
Counter({0: 250970, 1: 1833})
CPU times: user 1h 48min 34s, sys: 37.5 s, total: 1h 49min 12s
Wall time: 1h 49min 57s


%%time

from imblearn.under_sampling import AllKNN 

print(Counter(y_train))
X_AllKNN, y_AllKNN = AllKNN().fit_resample(X_train, y_train)
print(Counter(y_AllKNN))

Counter({0: 255560, 1: 1833})
Counter({0: 252250, 1: 1833})
CPU times: user 10min 18s, sys: 3.78 s, total: 10min 22s
Wall time: 10min 25s


%%time

from imblearn.under_sampling import InstanceHardnessThreshold

print(Counter(y_train))
X_IHT, y_IHT = InstanceHardnessThreshold(random_state=1).fit_resample(X_train, y_train)
print(Counter(y_IHT))

Counter({0: 255560, 1: 1833})
Counter({0: 237525, 1: 1833})
CPU times: user 1min 56s, sys: 920 ms, total: 1min 57s
Wall time: 1min 58s


%%time

print(Counter(y_train))
undersample_OneSidedSelection = OneSidedSelection(n_neighbors=1, n_seeds_S=200)
X_OSS, y_OSS = undersample_OneSidedSelection.fit_resample(X_train, y_train)
print(Counter(y_OSS))
'''
X_OSS.to_pickle("./X_OSS.pkl")  
y_OSS.to_pickle("./y_OSS.pkl")
'''

Counter({0: 255560, 1: 1833})
Counter({0: 89636, 1: 1833})
CPU times: user 27.6 s, sys: 289 ms, total: 27.9 s
Wall time: 27.8 s

'\nX_OSS.to_pickle("./X_OSS.pkl")  \ny_OSS.to_pickle("./y_OSS.pkl")\n'


%%time

print(Counter(y_train))
undersample_NCR = NeighbourhoodCleaningRule(sampling_strategy='majority', n_neighbors=3, threshold_cleaning=0.5)
X_NCR, y_NCR = undersample_NCR.fit_resample(X_train, y_train)
print(Counter(y_NCR))
'''
X_NCR.to_pickle("./X_NCR.pkl")  
y_NCR.to_pickle("./y_NCR.pkl")
'''

Counter({0: 255560, 1: 1833})
Counter({0: 252276, 1: 1833})
CPU times: user 5min 35s, sys: 3.65 s, total: 5min 39s
Wall time: 5min 58s

'\nX_NCR.to_pickle("./X_NCR.pkl")  \ny_NCR.to_pickle("./y_NCR.pkl")\n'


dill.dump_session('TFM_notebook_env.db')


%%time

print(Counter(y_train))
resample_SMOTETomek = SMOTETomek()
X_SMOTETomek, y_SMOTETomek = resample_SMOTETomek.fit_resample(X_train, y_train)
print(Counter(y_SMOTETomek))
print('\n')
'''
X_SMOTETomek.to_pickle("./X_SMOTETomek.pkl")  
y_SMOTETomek.to_pickle("./y_SMOTETomek.pkl")
'''

Counter({0: 255560, 1: 1833})
Counter({0: 255491, 1: 255491})


CPU times: user 8min 21s, sys: 5.74 s, total: 8min 27s
Wall time: 9min 33s

'\nX_SMOTETomek.to_pickle("./X_SMOTETomek.pkl")  \ny_SMOTETomek.to_pickle("./y_SMOTETomek.pkl")\n'


%%time

print(Counter(y_train))
resample_SMOTETomek_majority = SMOTETomek(tomek=TomekLinks(sampling_strategy='majority'))
X_SMOTETomekM, y_SMOTETomekM = resample_SMOTETomek_majority.fit_resample(X_train, y_train)
print(Counter(y_SMOTETomekM))
print('\n')
'''
X_SMOTETomek.to_pickle("./X_SMOTETomek.pkl")  
y_SMOTETomek.to_pickle("./y_SMOTETomek.pkl")
'''

Counter({0: 255560, 1: 1833})
Counter({1: 255560, 0: 255494})


CPU times: user 7min 21s, sys: 4.61 s, total: 7min 25s
Wall time: 7min 55s

'\nX_SMOTETomek.to_pickle("./X_SMOTETomek.pkl")  \ny_SMOTETomek.to_pickle("./y_SMOTETomek.pkl")\n'


%%time

print(Counter(y_train))
resample_SMOTEENN = SMOTEENN()
X_SMOTEENN, y_SMOTEENN = resample_SMOTEENN.fit_resample(X_train, y_train)
print(Counter(y_SMOTEENN))
print('\n')
'''
X_SMOTEENN.to_pickle("./X_SMOTEENN.pkl")  
y_SMOTEENN.to_pickle("./y_SMOTEENN.pkl")
'''

Counter({0: 255560, 1: 1833})
Counter({1: 252266, 0: 247119})


CPU times: user 5min 52s, sys: 3.35 s, total: 5min 55s
Wall time: 6min 13s

'\nX_SMOTEENN.to_pickle("./X_SMOTEENN.pkl")  \ny_SMOTEENN.to_pickle("./y_SMOTEENN.pkl")\n'


%%time

print(Counter(y_train))
resample_SMOTEENN_majority = SMOTEENN(enn=EditedNearestNeighbours(sampling_strategy='majority'))
X_SMOTEENNM, y_SMOTEENNM = resample_SMOTEENN_majority.fit_resample(X_train, y_train)
print(Counter(y_SMOTEENNM))
print('\n')
'''
X_SMOTEENN.to_pickle("./X_SMOTEENN.pkl")  
y_SMOTEENN.to_pickle("./y_SMOTEENN.pkl")
'''

Counter({0: 255560, 1: 1833})
Counter({1: 255560, 0: 247250})


CPU times: user 5min 39s, sys: 3.5 s, total: 5min 43s
Wall time: 6min 3s

'\nX_SMOTEENN.to_pickle("./X_SMOTEENN.pkl")  \ny_SMOTEENN.to_pickle("./y_SMOTEENN.pkl")\n'


dill.dump_session('TFM_notebook_env.db')


#import dill
#dill.load_session('/content/drive/MyDrive/Colab/TFM_notebook_env.db')
#dill.load_session('TFM_notebook_env.db')


from sklearn.metrics import accuracy_score, f1_score, confusion_matrix, recall_score

dummy_clf = DummyClassifier(strategy="most_frequent")
dummy_clf.fit(X_train, y_train)

#print(dummy_clf.score(X_val, y_val))

pred = dummy_clf.predict(X_test)
print("Precisión: {:.3f}".format(accuracy_score(y_test, pred))) 
print("F1-binary: {:.3f}".format(f1_score(y_test, pred, average='binary'))) 
print("Recall: {:.3f}".format(recall_score(y_test, pred))) 


matriz = confusion_matrix(y_test, pred)

sns.heatmap(matriz, annot=True, fmt="d", cmap='Blues', square=True)
plt.xlabel("predicción")
plt.ylabel("real")
plt.title('Matriz de confusión', y=1.05, size=18)
plt.show()

Precisión: 0.993
F1-binary: 0.000
Recall: 0.000


from sklearn.linear_model import LogisticRegression

def basic_balancedLRC(trainX, trainy, testX, testy, weight='balanced'):
    
    balancedLrc = LogisticRegression(random_state=0, solver='lbfgs', class_weight='balanced')
    balancedLrc.fit(trainX, trainy)
    
    pred = balancedLrc.predict(testX)
    print("Accuracy: {:.3f}".format(accuracy_score(testy, pred)))
    print("F1-binary: {:.3f}".format(f1_score(testy, pred, average='binary'))) 
    print("Recall: {:.3f}".format(recall_score(testy, pred))) 

    matriz = confusion_matrix(testy, pred)
    sns.heatmap(matriz, annot=True, fmt="d", cmap='Blues', square=True)
    plt.xlabel("predicción")
    plt.ylabel("real")
    plt.title('Matriz de confusión', y=1.05, size=18)
    plt.show()
    print('\n')


%%time
basic_balancedLRC(trainX=X_train, trainy=y_train, testX=X_test, testy=y_test, weight='balanced')

/Users/inigoab/opt/anaconda3/envs/TFM/lib/python3.7/site-packages/sklearn/linear_model/_logistic.py:765: ConvergenceWarning: lbfgs failed to converge (status=1):
STOP: TOTAL NO. of ITERATIONS REACHED LIMIT.

Increase the number of iterations (max_iter) or scale the data as shown in:
    https://scikit-learn.org/stable/modules/preprocessing.html
Please also refer to the documentation for alternative solver options:
    https://scikit-learn.org/stable/modules/linear_model.html#logistic-regression
  extra_warning_msg=_LOGISTIC_SOLVER_CONVERGENCE_MSG)

Accuracy: 0.918
F1-binary: 0.131
Recall: 0.870


CPU times: user 13.1 s, sys: 576 ms, total: 13.6 s
Wall time: 3.68 s


def grid_LRC(trainX, trainy, testX, testy, balance):
    
    param_grid = dict(class_weight=balance)
    
    cv = RepeatedStratifiedKFold(n_splits=10, n_repeats=2, random_state=1)
    
    grid_model = GridSearchCV(estimator=LogisticRegression(random_state=0, solver='lbfgs'),
                              param_grid=param_grid, n_jobs=-1, cv=cv, scoring='recall')
    
    grid_model.fit(trainX, trainy)
                              
    print("\nEl mejor parámetros es: {}".format(grid_model.best_params_))
    
    best_class_weight = grid_model.best_params_['class_weight']
    
    balancedLrc = LogisticRegression(class_weight=best_class_weight, 
                                         random_state=0, verbose=1)
    
    balancedLrc.fit(trainX, trainy)
    
    pred = balancedLrc.predict(testX)
    print("Accuracy: {:.3f}".format(accuracy_score(testy, pred)))
    print("F1-binary: {:.3f}".format(f1_score(testy, pred, average='binary'))) 
    print("Recall: {:.3f}".format(recall_score(testy, pred))) 

    matriz = confusion_matrix(testy, pred)
    sns.heatmap(matriz, annot=True, fmt="d", cmap='Blues', square=True)
    plt.xlabel("predicción")
    plt.ylabel("real")
    plt.title('Matriz de confusión', y=1.05, size=18)
    plt.show()
    print('\n')


from sklearn.naive_bayes import GaussianNB

def basic_GNB(trainX, trainy, testX, testy):
    
    GNBC = GaussianNB()
    GNBC.fit(trainX, trainy)
    
    pred = GNBC.predict(testX)
    print("Accuracy: {:.3f}".format(accuracy_score(testy, pred)))
    print("F1-binary: {:.3f}".format(f1_score(testy, pred, average='binary'))) 
    print("Recall: {:.3f}".format(recall_score(testy, pred))) 

    matriz = confusion_matrix(testy, pred)
    sns.heatmap(matriz, annot=True, fmt="d", cmap='Blues', square=True)
    plt.xlabel("predicción")
    plt.ylabel("real")
    plt.title('Matriz de confusión', y=1.05, size=18)
    plt.show()
    print('\n')


%%time
basic_GNB(trainX=X_train, trainy=y_train, testX=X_test, testy=y_test)

Accuracy: 0.961
F1-binary: 0.086
Recall: 0.254


CPU times: user 858 ms, sys: 70.4 ms, total: 929 ms
Wall time: 620 ms


from sklearn.neighbors import KNeighborsClassifier

def basic_KNN(trainX, trainy, testX, testy):
    
    KNNC = KNeighborsClassifier(n_neighbors=3)
    KNNC.fit(trainX, trainy)
    
    pred = KNNC.predict(testX)
    print("Accuracy: {:.3f}".format(accuracy_score(testy, pred)))
    print("F1-binary: {:.3f}".format(f1_score(testy, pred, average='binary'))) 
    print("Recall: {:.3f}".format(recall_score(testy, pred))) 

    matriz = confusion_matrix(testy, pred)
    sns.heatmap(matriz, annot=True, fmt="d", cmap='Blues', square=True)
    plt.xlabel("predicción")
    plt.ylabel("real")
    plt.title('Matriz de confusión', y=1.05, size=18)
    plt.show()
    print('\n')


%%time
basic_KNN(X_train, y_train, X_test, y_test)

Accuracy: 0.993
F1-binary: 0.337
Recall: 0.247


CPU times: user 2min 3s, sys: 1.26 s, total: 2min 4s
Wall time: 2min 8s


def basic_balancedDTC(trainX, trainy, testX, testy, weight='balanced'):
    
    balancedDTC = DecisionTreeClassifier(random_state=0, class_weight='balanced')
    balancedDTC.fit(trainX, trainy)
    
    pred = balancedDTC.predict(testX)
    print("Accuracy: {:.3f}".format(accuracy_score(testy, pred)))
    print("F1-binary: {:.3f}".format(f1_score(testy, pred, average='binary'))) 
    print("Recall: {:.3f}".format(recall_score(testy, pred))) 

    matriz = confusion_matrix(testy, pred)
    sns.heatmap(matriz, annot=True, fmt="d", cmap='Blues', square=True)
    plt.xlabel("predicción")
    plt.ylabel("real")
    plt.title('Matriz de confusión', y=1.05, size=18)
    plt.show()
    print('\n')


%%time
basic_balancedDTC(trainX=X_train, trainy=y_train, testX=X_test, testy=y_test, weight='balanced')

Accuracy: 0.981
F1-binary: 0.216
Recall: 0.369


CPU times: user 1.58 s, sys: 58.7 ms, total: 1.64 s
Wall time: 1.33 s


def grid_balancedDTC(trainX, trainy, testX, testy, 
                     max_depth= range(4, 11),
                     min_samples_split= [2, 10, 20, 50, 100],
                     class_weight= [{0:1,1:1}, {0:1,1:10}, {0:1,1:100}],
                     scoring='recall'):
    
    param_grid = {"max_depth": max_depth, 
                  "min_samples_split": min_samples_split,
                  "class_weight": class_weight}
    
    cv = RepeatedStratifiedKFold(n_splits=10, n_repeats=3, random_state=1)
    
    grid_model = GridSearchCV(DecisionTreeClassifier(random_state=0), param_grid=param_grid,
                                    cv=cv, n_jobs= -1, scoring=scoring, verbose=1)
    grid_model.fit(trainX, trainy)
    
    print("\nLos mejores parámetros son: {}".format(grid_model.best_params_))
    
    best_max_depth = grid_model.best_params_['max_depth']
    best_min_samples_split = grid_model.best_params_['min_samples_split']
    best_class_weight = grid_model.best_params_['class_weight']
    
    grid_dtc = DecisionTreeClassifier(max_depth=best_max_depth, 
                                 min_samples_split=best_min_samples_split, 
                                 random_state=20)
    
    grid_dtc.fit(trainX, trainy)
    
    pred = grid_dtc.predict(testX)
    print("Accuracy: {:.3f}".format(accuracy_score(testy, pred)))
    print("F1-binary: {:.3f}".format(f1_score(testy, pred, average='binary'))) 
    print("Recall: {:.3f}".format(recall_score(testy, pred))) 

    matriz = confusion_matrix(testy, pred)
    sns.heatmap(matriz, annot=True, fmt="d", cmap='Blues', square=True)
    plt.xlabel("predicción")
    plt.ylabel("real")
    plt.title('Matriz de confusión', y=1.05, size=18)
    plt.show()
    print('\n')


%%time

grid_balancedDTC(trainX=X_train, trainy=y_train, testX=X_test, testy=y_test, 
                 max_depth= range(9, 11),
                 min_samples_split= [2, 10],
                 class_weight= [{0:1,1:10},{0:1,1:100}],
                 scoring= 'recall')

Fitting 30 folds for each of 8 candidates, totalling 240 fits

Los mejores parámetros son: {'class_weight': {0: 1, 1: 100}, 'max_depth': 9, 'min_samples_split': 10}
Accuracy: 0.994
F1-binary: 0.295
Recall: 0.188


CPU times: user 5.18 s, sys: 1.34 s, total: 6.52 s
Wall time: 2min 12s


from sklearn.svm import SVC

def basic_balancedSVC(trainX, trainy, testX, testy, weight='balanced'):
    
    balancedSVC = SVC(gamma='scale', class_weight='balanced')    
    balancedSVC.fit(trainX, trainy)
    
    pred = balancedSVC.predict(testX)
    print("Accuracy: {:.3f}".format(accuracy_score(testy, pred)))
    print("F1-binary: {:.3f}".format(f1_score(testy, pred, average='binary'))) 
    print("Recall: {:.3f}".format(recall_score(testy, pred))) 

    matriz = confusion_matrix(testy, pred)
    sns.heatmap(matriz, annot=True, fmt="d", cmap='Blues', square=True)
    plt.xlabel("predicción")
    plt.ylabel("real")
    plt.title('Matriz de confusión', y=1.05, size=18)
    plt.show()
    print('\n')


'''
%%time
basic_balancedSVC(trainX=X_train, trainy=y_train, testX=X_test, testy=y_test, y_test, weight='balanced')
'''

"\n%%time\nbasic_balancedSVC(trainX=X_train, trainy=y_train, testX=X_test, testy=y_test, y_test, weight='balanced')\n"


def balancedXGB(trainX, trainy, testX, testy, weight='balanced', eval_metric='aucpr'):
    
    from sklearn.model_selection import cross_val_predict

    if weight == False or weight == None:
        estimate = 1.0
    elif weight == 'balanced':
        counter = Counter(trainy)
        estimate = counter[0]/counter[1]
        print('Peso estimado: ', estimate)
    else:
        estimate = weight
  
    xgb = XGBClassifier(scale_pos_weight=estimate, use_label_encoder=False, eval_metric=eval_metric)
    xgb.fit(trainX, trainy)
  
    pred = xgb.predict(testX)
    print("Accuracy: {:.3f}".format(accuracy_score(testy, pred)))
    print("F1-binary: {:.3f}".format(f1_score(testy, pred, average='binary'))) 
    print("Recall: {:.3f}".format(recall_score(testy, pred))) 
  
    matriz = confusion_matrix(testy, pred)
    sns.heatmap(matriz, annot=True, fmt="d", cmap='Blues', square=True)
    plt.xlabel("predicción")
    plt.ylabel("real")
    plt.title('Matriz de confusión', y=1.05, size=18)
    plt.show()


%%time
balancedXGB(trainX=X_train, trainy=y_train, testX=X_test, testy=y_test, weight=False)

Accuracy: 0.994
F1-binary: 0.412
Recall: 0.298

CPU times: user 1min 5s, sys: 1.65 s, total: 1min 7s
Wall time: 26.1 s


%%time
balancedXGB(trainX=X_train, trainy=y_train, testX=X_test, testy=y_test, weight='balanced')

Peso estimado:  139.4217130387343
Accuracy: 0.962
F1-binary: 0.240
Recall: 0.845

CPU times: user 1min 9s, sys: 1.69 s, total: 1min 10s
Wall time: 24.6 s


%%time
balancedXGB(trainX=X_train0, trainy=y_train, testX=X_test0, testy=y_test, weight='balanced')

Peso estimado:  139.4217130387343
Accuracy: 0.966
F1-binary: 0.260
Recall: 0.830

CPU times: user 2min 33s, sys: 2.44 s, total: 2min 36s
Wall time: 58.4 s


def basic_balancedRFC(trainX, trainy, testX, testy, weight='balanced'):

    balancedRfc = RandomForestClassifier(random_state=20, class_weight=weight)
    balancedRfc.fit(trainX, trainy)
    
    pred = balancedRfc.predict(testX)
    print("Accuracy: {:.3f}".format(accuracy_score(testy, pred)))
    print("F1-binary: {:.3f}".format(f1_score(testy, pred, average='binary'))) 
    print("Recall: {:.3f}".format(recall_score(testy, pred))) 

    matriz = confusion_matrix(testy, pred)
    sns.heatmap(matriz, annot=True, fmt="d", cmap='Blues', square=True)
    plt.xlabel("predicción")
    plt.ylabel("real")
    plt.title('Matriz de confusión', y=1.05, size=18)
    plt.show()
    print('\n')


%%time
basic_balancedRFC(trainX=X_train, trainy=y_train, testX=X_test, testy=y_test, weight='balanced')

Accuracy: 0.986
F1-binary: 0.239
Recall: 0.300


CPU times: user 31.6 s, sys: 746 ms, total: 32.3 s
Wall time: 35 s


def balancedRFC(trainX, trainy, testX, testy, 
                scoring='recall', weight='balanced',
                n_estimators=[50, 100, 200], max_depth=range(5,15),
                standardize=False, normalize=False, pca=False):
  
    steps = list()
  
    if standardize == True: steps.append(('standarizer', StandardScaler()))
  
    if normalize == True: steps.append(('normalizer', MinMaxScaler()))

    if not isinstance(pca, bool):
        if pca > 1: 
            steps.append(('pca', PCA(n_components=pca)))
            print('\nSe aplica PCA para reducir la dimensionalidad a {} compotentes.'.format(pca))
    elif pca:
        steps.append(('pca', PCA()))
  
    rfc = RandomForestClassifier(random_state=42, class_weight=weight)
    steps.append(('rfc', rfc))
    pipeline = Pipeline(steps=steps)
    
    param_grid = {"rfc__n_estimators": n_estimators, "rfc__max_depth": max_depth}

    cv = RepeatedStratifiedKFold(n_splits=10, n_repeats=2, random_state=1)
    
    grid_model = GridSearchCV(pipeline, 
                            param_grid, cv=cv, n_jobs= -1, scoring=scoring, verbose=1)
    
    grid_model.fit(trainX, trainy)
    
    print("\nLos mejores parámetros son: {}".format(grid_model.best_params_))
    
    best_n_estimators = grid_model.best_params_['rfc__n_estimators']
    best_max_depth = grid_model.best_params_['rfc__max_depth']
    
    balancedRfc = RandomForestClassifier(n_estimators=best_n_estimators, 
                                         max_depth=best_max_depth, 
                                         random_state=20, class_weight='balanced', verbose=1)
    
    balancedRfc.fit(trainX, trainy)
    
    pred = balancedRfc.predict(testX)
    print("\nAccuracy: {:.3f}".format(accuracy_score(testy, pred)))
    print("F1-binary: {:.3f}".format(f1_score(testy, pred, average='binary'))) 
    print("Recall: {:.3f}".format(recall_score(testy, pred))) 
    
    matriz = confusion_matrix(testy, pred)
    sns.heatmap(matriz, annot=True, fmt="d", cmap='Blues', square=True)
    plt.xlabel("predicción")
    plt.ylabel("real")
    plt.title('Matriz de confusión', y=1.05, size=18)
    plt.show()
    print('\n')


balancedRFC(trainX=X_train, trainy=y_train, testX=X_test, testy=y_test,
                scoring='recall', weight='balanced',
                n_estimators=[50, 100, 200], max_depth=range(5,15),
                standardize=False, normalize=False, pca=False)

Fitting 20 folds for each of 30 candidates, totalling 600 fits

Los mejores parámetros son: {'rfc__max_depth': 5, 'rfc__n_estimators': 50}

[Parallel(n_jobs=1)]: Using backend SequentialBackend with 1 concurrent workers.
[Parallel(n_jobs=1)]: Done  50 out of  50 | elapsed:    9.0s finished
[Parallel(n_jobs=1)]: Using backend SequentialBackend with 1 concurrent workers.
[Parallel(n_jobs=1)]: Done  50 out of  50 | elapsed:    0.3s finished

Accuracy: 0.878
F1-binary: 0.101
Recall: 0.957


def grid_balancedRFC(trainX, trainy, testX, testy, 
                scoring='recall', weights=[{0:1,1:5}, {0:1,1:10}, {0:1,1:20}],
                n_estimators=[50, 100, 200], max_depth=range(5,15),
                standardize=False, normalize=False, pca=False):
  
    steps = list()
  
    if standardize == True: steps.append(('standarizer', StandardScaler()))
  
    if normalize == True: steps.append(('normalizer', MinMaxScaler()))

    if not isinstance(pca, bool):
        if pca > 1: 
            steps.append(('pca', PCA(n_components=pca)))
            print('\nSe aplica PCA para reducir la dimensionalidad a {} compotentes.'.format(pca))
    elif pca:
        steps.append(('pca', PCA()))
  
    rfc = RandomForestClassifier(random_state=20, class_weight=weights)
    steps.append(('rfc', rfc))
    pipeline = Pipeline(steps=steps)
    
    param_grid = {"rfc__n_estimators": n_estimators, 
                  "rfc__max_depth": max_depth, 
                  "rfc__class_weight": balance}

    cv = RepeatedStratifiedKFold(n_splits=10, n_repeats=2, random_state=1)
    
    grid_model = GridSearchCV(pipeline, 
                            param_grid, cv=cv, n_jobs= -1, scoring=scoring, verbose=1)
    
    grid_model.fit(trainX, trainy)
    
    print("\nLos mejores parámetros son: {}".format(grid_model.best_params_))
    
    best_n_estimators = grid_model.best_params_['rfc__n_estimators']
    best_max_depth = grid_model.best_params_['rfc__max_depth']
    best_class_weight = grid_model.best_params_['rfc__class_weight']

    
    grid_balancedRfc = RandomForestClassifier(n_estimators=best_n_estimators, 
                                         max_depth=best_max_depth, 
                                         class_weight=best_class_weight,
                                         random_state=20, verbose=1)
    
    grid_balancedRfc.fit(trainX, trainy)
    
    pred = grid_balancedRfc.predict(testX)
    print("\nAccuracy: {:.3f}".format(accuracy_score(testy, pred)))
    print("F1-binary: {:.3f}".format(f1_score(testy, pred, average='binary'))) 
    print("Recall: {:.3f}".format(recall_score(testy, pred))) 
    
    matriz = confusion_matrix(testy, pred)
    sns.heatmap(matriz, annot=True, fmt="d", cmap='Blues', square=True)
    plt.xlabel("predicción")
    plt.ylabel("real")
    plt.title('Matriz de confusión', y=1.05, size=18)
    plt.show()
    print('\n')


dill.dump_session('TFM_notebook_env.db')


#import dill
#dill.load_session('/content/drive/MyDrive/Colab/TFM_notebook_env.db')
#dill.load_session('TFM_notebook_env.db')


%whos

Variable                          Type                         Data/Info
------------------------------------------------------------------------
ADASYN                            ABCMeta                      <class 'imblearn.over_sampling._adasyn.ADASYN'>
AllKNN                            ABCMeta                      <class 'imblearn.under_sa<...>arest_neighbours.AllKNN'>
BorderlineSMOTE                   ABCMeta                      <class 'imblearn.over_sam<...>._smote.BorderlineSMOTE'>
ClusterCentroids                  ABCMeta                      <class 'imblearn.under_sa<...>troids.ClusterCentroids'>
CondensedNearestNeighbour         ABCMeta                      <class 'imblearn.under_sa<...>ndensedNearestNeighbour'>
Counter                           type                         <class 'collections.Counter'>
DF                                DataFrame                           TIPOUSUARIO     CA<...>367705 rows x 11 columns]
DecisionTreeClassifier            ABCMeta                      <class 'sklearn.tree._cla<...>.DecisionTreeClassifier'>
DummyClassifier                   type                         <class 'sklearn.dummy.DummyClassifier'>
EditedNearestNeighbours           ABCMeta                      <class 'imblearn.under_sa<...>EditedNearestNeighbours'>
GaussianNB                        ABCMeta                      <class 'sklearn.naive_bayes.GaussianNB'>
GridSearchCV                      ABCMeta                      <class 'sklearn.model_sel<...>on._search.GridSearchCV'>
InstanceHardnessThreshold         ABCMeta                      <class 'imblearn.under_sa<...>stanceHardnessThreshold'>
KNeighborsClassifier              ABCMeta                      <class 'sklearn.neighbors<...>on.KNeighborsClassifier'>
LogisticRegression                type                         <class 'sklearn.linear_mo<...>stic.LogisticRegression'>
MinMaxScaler                      type                         <class 'sklearn.preproces<...>sing._data.MinMaxScaler'>
NearMiss                          ABCMeta                      <class 'imblearn.under_sa<...>tion._nearmiss.NearMiss'>
NeighbourhoodCleaningRule         ABCMeta                      <class 'imblearn.under_sa<...>ighbourhoodCleaningRule'>
OneHotEncoder                     type                         <class 'sklearn.preproces<...>_encoders.OneHotEncoder'>
OneSidedSelection                 ABCMeta                      <class 'imblearn.under_sa<...>ction.OneSidedSelection'>
OrdinalEncoder                    type                         <class 'sklearn.preproces<...>encoders.OrdinalEncoder'>
PCA                               ABCMeta                      <class 'sklearn.decomposition._pca.PCA'>
Pipeline                          ABCMeta                      <class 'sklearn.pipeline.Pipeline'>
RFECV                             ABCMeta                      <class 'sklearn.feature_selection._rfe.RFECV'>
RandomForestClassifier            ABCMeta                      <class 'sklearn.ensemble.<...>.RandomForestClassifier'>
RandomOverSampler                 ABCMeta                      <class 'imblearn.over_sam<...>mpler.RandomOverSampler'>
RandomUnderSampler                ABCMeta                      <class 'imblearn.under_sa<...>pler.RandomUnderSampler'>
RepeatedEditedNearestNeighbours   ABCMeta                      <class 'imblearn.under_sa<...>EditedNearestNeighbours'>
RepeatedStratifiedKFold           ABCMeta                      <class 'sklearn.model_sel<...>RepeatedStratifiedKFold'>
SMOTE                             ABCMeta                      <class 'imblearn.over_sampling._smote.SMOTE'>
SMOTEENN                          ABCMeta                      <class 'imblearn.combine._smote_enn.SMOTEENN'>
SMOTENC                           ABCMeta                      <class 'imblearn.over_sampling._smote.SMOTENC'>
SMOTETomek                        ABCMeta                      <class 'imblearn.combine.<...>_smote_tomek.SMOTETomek'>
SVC                               ABCMeta                      <class 'sklearn.svm._classes.SVC'>
SVMSMOTE                          ABCMeta                      <class 'imblearn.over_sampling._smote.SVMSMOTE'>
SelectKBest                       ABCMeta                      <class 'sklearn.feature_s<...>e_selection.SelectKBest'>
StandardScaler                    type                         <class 'sklearn.preproces<...>ng._data.StandardScaler'>
TomekLinks                        ABCMeta                      <class 'imblearn.under_sa<...>_tomek_links.TomekLinks'>
X                                 DataFrame                           TIPOUSUARIO       <...>110312 rows x 10 columns]
XGBClassifier                     type                         <class 'xgboost.sklearn.XGBClassifier'>
X_ADASYN                          DataFrame                                  PF      SPA<...>[511489 rows x 9 columns]
X_AllKNN                          DataFrame                             PF  SPAM  email_<...>[254083 rows x 9 columns]
X_BS                              DataFrame                                  PF      SPA<...>[511120 rows x 9 columns]
X_CC                              DataFrame                                PF          S<...>\n[3666 rows x 9 columns]
X_ENN                             DataFrame                             PF  SPAM  email_<...>[254856 rows x 9 columns]
X_IHT                             DataFrame                             PF  SPAM  email_<...>[239358 rows x 9 columns]
X_NCR                             DataFrame                             PF  SPAM  email_<...>[254109 rows x 9 columns]
X_NM1                             DataFrame                           PF  SPAM  email_es<...>\n[3666 rows x 9 columns]
X_NM3                             DataFrame                           PF  SPAM  email_es<...>\n[3666 rows x 9 columns]
X_OSS                             DataFrame                            PF  SPAM  email_e<...>n[91469 rows x 9 columns]
X_RENN                            DataFrame                             PF  SPAM  email_<...>[252803 rows x 9 columns]
X_ROS                             DataFrame                             PF  SPAM  email_<...>[511120 rows x 9 columns]
X_RUS                             DataFrame                           PF  SPAM  email_es<...>\n[3666 rows x 9 columns]
X_SMOTE                           DataFrame                                  PF  SPAM  e<...>[511120 rows x 9 columns]
X_SMOTEENN                        DataFrame                                  PF      SPA<...>[499385 rows x 9 columns]
X_SMOTEENNM                       DataFrame                                  PF      SPA<...>[502810 rows x 9 columns]
X_SMOTENC                         DataFrame                             PF  SPAM  email_<...>[511120 rows x 9 columns]
X_SMOTETomek                      DataFrame                             PF      SPAM  em<...>[510982 rows x 9 columns]
X_SMOTETomekM                     DataFrame                                  PF      SPA<...>[511054 rows x 9 columns]
X_SVMSMOTE                        DataFrame                                  PF      SPA<...>[511120 rows x 9 columns]
X_TL                              DataFrame                             PF  SPAM  email_<...>[257035 rows x 9 columns]
X_test                            DataFrame                            PF  SPAM  email_e<...>n[55156 rows x 9 columns]
X_test0                           DataFrame                           No_email   PF   PJ<...>[55156 rows x 30 columns]
X_test1                           DataFrame                           TIPOUSUARIO       <...>110312 rows x 10 columns]
X_train                           DataFrame                             PF  SPAM  email_<...>[257393 rows x 9 columns]
X_train0                          DataFrame                            No_email   PF   P<...>257393 rows x 30 columns]
X_val                             DataFrame                            PF  SPAM  email_e<...>n[55156 rows x 9 columns]
X_val0                            DataFrame                           No_email   PF   PJ<...>[55156 rows x 30 columns]
accuracy_score                    function                     <function accuracy_score at 0x11d062290>
ax                                AxesSubplot                  AxesSubplot(0.125,0.125;0.775x0.755)
balancedRFC                       function                     <function balancedRFC at 0x153cd7560>
balancedXGB                       function                     <function balancedXGB at 0x152a6a8c0>
basic_GNB                         function                     <function basic_GNB at 0x16f2a6ef0>
basic_KNN                         function                     <function basic_KNN at 0x153cd5170>
basic_balancedDTC                 function                     <function basic_balancedDTC at 0x152b92710>
basic_balancedLRC                 function                     <function basic_balancedLRC at 0x152a6a050>
basic_balancedRFC                 function                     <function basic_balancedRFC at 0x152df8c20>
basic_balancedSVC                 function                     <function basic_balancedSVC at 0x152a6a7a0>
chi2                              function                     <function chi2 at 0x11d231200>
chi2_contingency                  function                     <function chi2_contingency at 0x11c6c0200>
confusion_matrix                  function                     <function confusion_matrix at 0x11d062440>
consumosDF                        DataFrame                           IDCONSUMO IDUSUARI<...>796020 rows x 10 columns]
consumos_agrupados                DataFrame                    DESCPRODUCTO IDUSUARIO  F<...>[367638 rows x 3 columns]
consumos_fechas                   DataFrame                           IDUSUARIO primer_c<...>[367638 rows x 3 columns]
corr_matrix                       function                     <function corr_matrix at 0x152af9f80>
counter                           Counter                      Counter({0: 365086, 1: 2619})
dill                              module                       <module 'dill' from '/Use<...>ckages/dill/__init__.py'>
dummy_clf                         DummyClassifier              DummyClassifier(strategy='most_frequent')
f1_score                          function                     <function f1_score at 0x11d062e60>
frecuencias_email                 Series                       gmail.com           0.434<...>th: 15446, dtype: float64
grid_LRC                          function                     <function grid_LRC at 0x152b92f80>
grid_balancedDTC                  function                     <function grid_balancedDTC at 0x153ced830>
grid_balancedRFC                  function                     <function grid_balancedRFC at 0x21b8a9c20>
k                                 int64                        1
loginsDF                          DataFrame                           IDUSUARIO FECHALOG<...>[616894 rows x 3 columns]
logins_agrupados                  DataFrame                           IDUSUARIO  N_login<...>[197129 rows x 2 columns]
logins_fechas                     DataFrame                           IDUSUARIO primer_l<...>[197129 rows x 3 columns]
matriz                            ndarray                      2x2: 4 elems, type `int64`, 32 bytes
np                                module                       <module 'numpy' from '/Us<...>kages/numpy/__init__.py'>
ohe                               function                     <function ohe at 0x152b8d290>
p                                 Rectangle                    Rectangle(xy=(0.6, 0), wi<...>.8, height=2619, angle=0)
pd                                module                       <module 'pandas' from '/U<...>ages/pandas/__init__.py'>
per                               float                        0.7122557484940373
percentage                        str                          0.71%
plt                               module                       <module 'matplotlib.pyplo<...>es/matplotlib/pyplot.py'>
pred                              ndarray                      55156: 55156 elems, type `int64`, 441248 bytes (430.90625 kb)
r_conversion                      float                        0.007122557484940373
recall_score                      function                     <function recall_score at 0x11d069710>
resample_SMOTEENN                 SMOTEENN                     SMOTEENN()
resample_SMOTEENN_majority        SMOTEENN                     SMOTEENN(enn=EditedNeares<...>ing_strategy='majority'))
resample_SMOTETomek               SMOTETomek                   SMOTETomek()
resample_SMOTETomek_majority      SMOTETomek                   SMOTETomek(tomek=TomekLin<...>ing_strategy='majority'))
rfe                               RFECV                        RFECV(cv=5, estimator=Ran<...>te=20), scoring='recall')
sesionesDF                        DataFrame                           IDUSUARIO FECHA_SE<...>[407757 rows x 4 columns]
sesiones_agrupadas                DataFrame                           IDUSUARIO  N_sesio<...>[367702 rows x 2 columns]
sesiones_fechas                   DataFrame                           IDUSUARIO primera_<...>[367702 rows x 3 columns]
sns                               module                       <module 'seaborn' from '/<...>ges/seaborn/__init__.py'>
test_anova                        function                     <function test_anova at 0x152d82050>
test_chi2                         function                     <function test_chi2 at 0x152b3e9e0>
total                             int                          367705
train_test_split                  function                     <function train_test_split at 0x11cf637a0>
undersample_ENN                   EditedNearestNeighbours      EditedNearestNeighbours()
undersample_NCR                   NeighbourhoodCleaningRule    NeighbourhoodCleaningRule<...>ling_strategy='majority')
undersample_NearMiss1             NearMiss                     NearMiss()
undersample_NearMiss3             NearMiss                     NearMiss(version=3)
undersample_OneSidedSelection     OneSidedSelection            OneSidedSelection(n_neighbors=1, n_seeds_S=200)
undersample_RandomUnderSampler    RandomUnderSampler           RandomUnderSampler(random_state=1)
undersample_TL                    TomekLinks                   TomekLinks()
usuariosDF                        DataFrame                           IDUSUARIO TIPOUSUA<...>367705 rows x 19 columns]
v                                 int                          2619
x                                 float64                      0.95
y                                 Series                       49328     0\n347838    0\<...>gth: 110312, dtype: int64
y_ADASYN                          Series                       0         0\n1         0\<...>gth: 511489, dtype: int64
y_AllKNN                          Series                       0         0\n1         0\<...>gth: 254083, dtype: int64
y_BS                              Series                       0         0\n1         0\<...>gth: 511120, dtype: int64
y_CC                              Series                       0       0\n1       0\n2  <...>ength: 3666, dtype: int64
y_ENN                             Series                       0         0\n1         0\<...>gth: 254856, dtype: int64
y_IHT                             Series                       0         0\n1         0\<...>gth: 239358, dtype: int64
y_NCR                             Series                       0         0\n1         0\<...>gth: 254109, dtype: int64
y_NM1                             Series                       0       0\n1       0\n2  <...>ength: 3666, dtype: int64
y_NM3                             Series                       0       0\n1       0\n2  <...>ength: 3666, dtype: int64
y_OSS                             Series                       0        0\n1        0\n2<...>ngth: 91469, dtype: int64
y_RENN                            Series                       0         0\n1         0\<...>gth: 252803, dtype: int64
y_ROS                             Series                       0         0\n1         0\<...>gth: 511120, dtype: int64
y_RUS                             Series                       0       0\n1       0\n2  <...>ength: 3666, dtype: int64
y_SMOTE                           Series                       0         0\n1         0\<...>gth: 511120, dtype: int64
y_SMOTEENN                        Series                       0         0\n1         0\<...>gth: 499385, dtype: int64
y_SMOTEENNM                       Series                       0         0\n1         0\<...>gth: 502810, dtype: int64
y_SMOTENC                         Series                       0         0\n1         0\<...>gth: 511120, dtype: int64
y_SMOTETomek                      Series                       0         0\n1         0\<...>gth: 510982, dtype: int64
y_SMOTETomekM                     Series                       0         0\n1         0\<...>gth: 511054, dtype: int64
y_SVMSMOTE                        Series                       0         0\n1         0\<...>gth: 511120, dtype: int64
y_TL                              Series                       0         0\n1         0\<...>gth: 257035, dtype: int64
y_test                            Series                       269699    0\n158840    0\<...>ngth: 55156, dtype: int64
y_test1                           Series                       49328     0\n347838    0\<...>gth: 110312, dtype: int64
y_train                           Series                       261599    0\n313436    0\<...>gth: 257393, dtype: int64
y_val                             Series                       279291    0\n342773    0\<...>ngth: 55156, dtype: int64


%%time
basic_balancedRFC(trainX=X_train0, trainy=y_train, testX=X_test0, testy=y_test,
                  weight='balanced')

Accuracy: 0.991
F1-binary: 0.250
Recall: 0.211


CPU times: user 36.6 s, sys: 961 ms, total: 37.6 s
Wall time: 37.7 s


%%time
basic_balancedRFC(trainX=X_train, trainy=y_train, testX=X_test, testy=y_test,
                  weight='balanced')

Accuracy: 0.986
F1-binary: 0.239
Recall: 0.300


CPU times: user 29 s, sys: 595 ms, total: 29.6 s
Wall time: 30 s


(X_train.columns == X_ROS.columns)

array([ True,  True,  True,  True,  True,  True,  True,  True,  True])


X_ROS.columns

Index(['PF', 'SPAM', 'email_español', 'IPCASOS', 'Perfil Promocional',
       'N_logins', 'N_sesiones', 'díasEntreConsumos', 'díasEntreLogins'],
      dtype='object')


X_ROS.shape

(511120, 9)


X_test.shape

(55156, 9)


y_test.shape

(55156,)


%%time
basic_balancedRFC(trainX=X_ROS, trainy=y_ROS, testX=X_test, testy=y_test,
                  weight='balanced')

Accuracy: 0.982
F1-binary: 0.237
Recall: 0.384


CPU times: user 1min 50s, sys: 2.18 s, total: 1min 52s
Wall time: 2min 1s


%%time
basic_balancedRFC(trainX=X_SMOTE, trainy=y_SMOTE, testX=X_test, testy=y_test,
                  weight='balanced')

Accuracy: 0.986
F1-binary: 0.290
Recall: 0.410


CPU times: user 1min 39s, sys: 1.85 s, total: 1min 41s
Wall time: 1min 43s


%%time
basic_balancedRFC(trainX=X_SMOTENC, trainy=y_SMOTENC, testX=X_test, testy=y_test, 
                  weight='balanced')

Accuracy: 0.974
F1-binary: 0.265
Recall: 0.646


CPU times: user 1min 46s, sys: 1.74 s, total: 1min 47s
Wall time: 1min 48s


%%time
basic_balancedRFC(trainX=X_BS, trainy=y_BS, testX=X_test, testy=y_test,
                  weight='balanced')

Accuracy: 0.987
F1-binary: 0.288
Recall: 0.356


CPU times: user 1min 34s, sys: 1.78 s, total: 1min 35s
Wall time: 1min 36s


%%time
basic_balancedRFC(trainX=X_SVMSMOTE, trainy=y_SVMSMOTE, testX=X_test, testy=y_test, 
                  weight='balanced')

Accuracy: 0.988
F1-binary: 0.332
Recall: 0.412


CPU times: user 1min 37s, sys: 2.11 s, total: 1min 39s
Wall time: 1min 41s


%%time
basic_balancedRFC(trainX=X_ADASYN, trainy=y_ADASYN, testX=X_test, testy=y_test, 
                  weight='balanced')

Accuracy: 0.985
F1-binary: 0.268
Recall: 0.382


CPU times: user 1min 7s, sys: 1.63 s, total: 1min 8s
Wall time: 1min 9s


%%time
dill.dump_session('TFM_notebook_env.db')

CPU times: user 1min 34s, sys: 4.49 s, total: 1min 38s
Wall time: 1min 41s


%%time
basic_balancedRFC(trainX=X_RUS, trainy=y_RUS, testX=X_test, testy=y_test, 
                  weight='balanced')

Accuracy: 0.904
F1-binary: 0.125
Recall: 0.957


CPU times: user 1.82 s, sys: 48.5 ms, total: 1.87 s
Wall time: 1.49 s


%%time
basic_balancedRFC(trainX=X_NM1, trainy=y_NM1, testX=X_test, testy=y_test, 
                  weight='balanced')

Accuracy: 0.349
F1-binary: 0.019
Recall: 0.906


CPU times: user 1.82 s, sys: 36.8 ms, total: 1.86 s
Wall time: 1.47 s


%%time
basic_balancedRFC(trainX=X_NM3, trainy=y_NM3, testX=X_test, testy=y_test,
                  weight='balanced')

Accuracy: 0.325
F1-binary: 0.016
Recall: 0.784


CPU times: user 1.97 s, sys: 44.7 ms, total: 2.01 s
Wall time: 1.62 s


%%time
basic_balancedRFC(trainX=X_CC, trainy=y_CC, testX=X_test, testy=y_test, 
                  weight='balanced')

Accuracy: 0.100
F1-binary: 0.015
Recall: 0.990


CPU times: user 1.81 s, sys: 36.3 ms, total: 1.85 s
Wall time: 1.46 s


%%time
basic_balancedRFC(trainX=X_TL, trainy=y_TL, testX=X_test, testy=y_test, 
                  weight='balanced')

Accuracy: 0.986
F1-binary: 0.246
Recall: 0.316


CPU times: user 27.6 s, sys: 434 ms, total: 28 s
Wall time: 27.7 s


%%time
basic_balancedRFC(trainX=X_ENN, trainy=y_ENN, testX=X_test, testy=y_test,
                  weight='balanced')

Accuracy: 0.986
F1-binary: 0.303
Recall: 0.417


CPU times: user 26.6 s, sys: 690 ms, total: 27.3 s
Wall time: 27.2 s


%%time
basic_balancedRFC(trainX=X_RENN, trainy=y_RENN, testX=X_test, testy=y_test, 
                  weight='balanced')

Accuracy: 0.985
F1-binary: 0.328
Recall: 0.499


CPU times: user 29.2 s, sys: 433 ms, total: 29.6 s
Wall time: 30.1 s


%%time
basic_balancedRFC(trainX=X_AllKNN, trainy=y_AllKNN, testX=X_test, testy=y_test, 
                  weight='balanced')

Accuracy: 0.986
F1-binary: 0.305
Recall: 0.433


CPU times: user 26.5 s, sys: 685 ms, total: 27.1 s
Wall time: 27.3 s


%%time
basic_balancedRFC(trainX=X_IHT, trainy=y_IHT, testX=X_test, testy=y_test, 
                  weight='balanced')

Accuracy: 0.974
F1-binary: 0.298
Recall: 0.763


CPU times: user 23 s, sys: 602 ms, total: 23.6 s
Wall time: 23.3 s


%%time
basic_balancedRFC(trainX=X_OSS, trainy=y_OSS, testX=X_test, testy=y_test, 
                  weight='balanced')

Accuracy: 0.990
F1-binary: 0.286
Recall: 0.295


CPU times: user 8.59 s, sys: 202 ms, total: 8.79 s
Wall time: 8.81 s


%%time
basic_balancedRFC(trainX=X_NCR, trainy=y_NCR, testX=X_test, testy=y_test, 
                  weight='balanced')

Accuracy: 0.986
F1-binary: 0.301
Recall: 0.415


CPU times: user 27.5 s, sys: 690 ms, total: 28.2 s
Wall time: 28.3 s


%%time
dill.dump_session('TFM_notebook_env.db')

CPU times: user 1min 34s, sys: 4.17 s, total: 1min 39s
Wall time: 1min 41s


%%time
basic_balancedRFC(trainX=X_SMOTETomek, trainy=y_SMOTETomek, testX=X_test, testy=y_test, 
                  weight='balanced')

Accuracy: 0.985
F1-binary: 0.278
Recall: 0.394


CPU times: user 1min 36s, sys: 1.87 s, total: 1min 38s
Wall time: 1min 40s


%%time
basic_balancedRFC(trainX=X_SMOTETomekM, trainy=y_SMOTETomekM, testX=X_test, testy=y_test, 
                  weight='balanced')

Accuracy: 0.986
F1-binary: 0.281
Recall: 0.397


CPU times: user 1min 36s, sys: 1.81 s, total: 1min 38s
Wall time: 1min 39s


%%time
basic_balancedRFC(trainX=X_SMOTEENN, trainy=y_SMOTEENN, testX=X_test, testy=y_test, 
                  weight='balanced')

Accuracy: 0.984
F1-binary: 0.341
Recall: 0.598


CPU times: user 1min 35s, sys: 2.13 s, total: 1min 37s
Wall time: 1min 37s


%%time
basic_balancedRFC(trainX=X_SMOTEENNM, trainy=y_SMOTEENNM, testX=X_test, testy=y_test, 
                  weight='balanced')

Accuracy: 0.983
F1-binary: 0.337
Recall: 0.603


CPU times: user 1min 35s, sys: 1.81 s, total: 1min 37s
Wall time: 1min 38s


%%time
dill.dump_session('TFM_notebook_env.db')

CPU times: user 1min 36s, sys: 4.34 s, total: 1min 41s
Wall time: 1min 44s


#import dill
#dill.load_session('TFM_notebook_env.db')


%%time

from imblearn.pipeline import Pipeline

print(Counter(y_train))
svmsmote = SVMSMOTE(sampling_strategy=0.3, random_state=42)
rus = RandomUnderSampler(random_state=42)
pipeline = Pipeline([('svmsmote', svmsmote), ('rus', rus)])
X_SMVSMOTENCRUS, y_SVMSMOTENCRUS = pipeline.fit_resample(X_train, y_train)
print(Counter(y_SVMSMOTENCRUS))
print('\n')

Counter({0: 255560, 1: 1833})
Counter({0: 76668, 1: 76668})


CPU times: user 6min 36s, sys: 3.22 s, total: 6min 40s
Wall time: 6min 43s


%%time
basic_balancedRFC(trainX=X_SMVSMOTENCRUS, trainy=y_SVMSMOTENCRUS, testX=X_test, testy=y_test, 
                  weight='balanced')

Accuracy: 0.984
F1-binary: 0.371
Recall: 0.644


CPU times: user 17 s, sys: 229 ms, total: 17.3 s
Wall time: 17.2 s


%%time

from imblearn.pipeline import Pipeline

print(Counter(y_train))
svmsmote = SVMSMOTE(sampling_strategy=0.1, random_state=42)
rus = RandomUnderSampler(random_state=42)
pipeline = Pipeline([('svmsmote', svmsmote), ('rus', rus)])
X_SMVSMOTENCRUS, y_SVMSMOTENCRUS = pipeline.fit_resample(X_train, y_train)
print(Counter(y_SVMSMOTENCRUS))
print('\n')

Counter({0: 255560, 1: 1833})
Counter({0: 25556, 1: 25556})


CPU times: user 6min 40s, sys: 3.28 s, total: 6min 43s
Wall time: 6min 49s


%%time
basic_balancedRFC(trainX=X_SMVSMOTENCRUS, trainy=y_SVMSMOTENCRUS, testX=X_test, testy=y_test, 
                  weight='balanced')

Accuracy: 0.976
F1-binary: 0.308
Recall: 0.743


CPU times: user 5.62 s, sys: 110 ms, total: 5.73 s
Wall time: 5.39 s


%%time

from imblearn.pipeline import Pipeline

print(Counter(y_train))
svmsmote = SVMSMOTE(sampling_strategy=0.05, random_state=42)
rus = RandomUnderSampler(random_state=42)
pipeline = Pipeline([('svmsmote', svmsmote), ('rus', rus)])
X_SMVSMOTENCRUS, y_SVMSMOTENCRUS = pipeline.fit_resample(X_train, y_train)
print(Counter(y_SVMSMOTENCRUS))
print('\n')

Counter({0: 255560, 1: 1833})
Counter({0: 12778, 1: 12778})


CPU times: user 6min 36s, sys: 3.22 s, total: 6min 40s
Wall time: 6min 44s


%%time
basic_balancedRFC(trainX=X_SMVSMOTENCRUS, trainy=y_SVMSMOTENCRUS, testX=X_test, testy=y_test, 
                  weight='balanced')

Accuracy: 0.966
F1-binary: 0.254
Recall: 0.817


CPU times: user 3.53 s, sys: 67.7 ms, total: 3.6 s
Wall time: 3.22 s


%%time

from imblearn.pipeline import Pipeline

print(Counter(y_train))
svmsmote = SVMSMOTE(sampling_strategy=0.01, random_state=42)
rus = RandomUnderSampler(random_state=42)
pipeline = Pipeline([('svmsmote', svmsmote), ('rus', rus)])
X_SMVSMOTENCRUS, y_SVMSMOTENCRUS = pipeline.fit_resample(X_train, y_train)
print(Counter(y_SVMSMOTENCRUS))
print('\n')

Counter({0: 255560, 1: 1833})
Counter({0: 2555, 1: 2555})


CPU times: user 6min 37s, sys: 3.23 s, total: 6min 40s
Wall time: 6min 44s


%%time
basic_balancedRFC(trainX=X_SMVSMOTENCRUS, trainy=y_SVMSMOTENCRUS, testX=X_test, testy=y_test, 
                  weight='balanced')

Accuracy: 0.928
F1-binary: 0.155
Recall: 0.926


CPU times: user 1.91 s, sys: 39.8 ms, total: 1.95 s
Wall time: 1.56 s


%%time

from imblearn.pipeline import Pipeline

print(Counter(y_train))
svmsmote = SVMSMOTE(sampling_strategy=0.008, random_state=42)
rus = RandomUnderSampler(random_state=42)
pipeline = Pipeline([('svmsmote', svmsmote), ('rus', rus)])
X_SMVSMOTENCRUS, y_SVMSMOTENCRUS = pipeline.fit_resample(X_train, y_train)
print(Counter(y_SVMSMOTENCRUS))
print('\n')

Counter({0: 255560, 1: 1833})
Counter({0: 2044, 1: 2044})


CPU times: user 6min 43s, sys: 3.39 s, total: 6min 46s
Wall time: 6min 51s


%%time
basic_balancedRFC(trainX=X_SMVSMOTENCRUS, trainy=y_SVMSMOTENCRUS, testX=X_test, testy=y_test, 
                  weight='balanced')

Accuracy: 0.920
F1-binary: 0.144
Recall: 0.941


CPU times: user 1.9 s, sys: 48.1 ms, total: 1.95 s
Wall time: 1.59 s


%%time

from imblearn.pipeline import Pipeline

print(Counter(y_train))
svmsmote = SVMSMOTE(sampling_strategy=0.1, random_state=42)
iht = InstanceHardnessThreshold(random_state=42)
pipeline = Pipeline([('svmsmote', svmsmote), ('iht', iht)])
X_SMVSMOTEIHT, y_SVMSMOTEIHT = pipeline.fit_resample(X_train, y_train)
print(Counter(y_SVMSMOTEIHT))
print('\n')

Counter({0: 255560, 1: 1833})
Counter({0: 236000, 1: 25556})


CPU times: user 9min 13s, sys: 5.49 s, total: 9min 18s
Wall time: 9min 26s


%%time
basic_balancedRFC(trainX=X_SMVSMOTEIHT, trainy=y_SVMSMOTEIHT, testX=X_test, testy=y_test, 
                  weight='balanced')

Accuracy: 0.972
F1-binary: 0.286
Recall: 0.786


CPU times: user 27.3 s, sys: 622 ms, total: 27.9 s
Wall time: 27.8 s


%%time

from imblearn.pipeline import Pipeline

print(Counter(y_train))
svmsmote = SVMSMOTE(sampling_strategy=0.01, random_state=42)
iht = InstanceHardnessThreshold(random_state=42)
pipeline = Pipeline([('svmsmote', svmsmote), ('iht', iht)])
X_SMVSMOTEIHT, y_SVMSMOTEIHT = pipeline.fit_resample(X_train, y_train)
print(Counter(y_SVMSMOTEIHT))
print('\n')

Counter({0: 255560, 1: 1833})
Counter({0: 237714, 1: 2555})


CPU times: user 9min 5s, sys: 4.99 s, total: 9min 10s
Wall time: 9min 16s


%%time
basic_balancedRFC(trainX=X_SMVSMOTEIHT, trainy=y_SVMSMOTEIHT, testX=X_test, testy=y_test, 
                  weight='balanced')

Accuracy: 0.974
F1-binary: 0.297
Recall: 0.763


CPU times: user 23.4 s, sys: 583 ms, total: 24 s
Wall time: 23.9 s


%%time

from imblearn.pipeline import Pipeline

print(Counter(y_train))
svmsmote = SVMSMOTE(sampling_strategy=0.01, random_state=42)
iht = InstanceHardnessThreshold(random_state=42)
rus = RandomUnderSampler(random_state=42)
pipeline = Pipeline([('svmsmote', svmsmote), ('iht', iht), ('rus', rus)])
X_SMVSMOTEIHTRUS, y_SVMSMOTEIHTRUS = pipeline.fit_resample(X_train, y_train)
print(Counter(y_SVMSMOTEIHTRUS))
print('\n')

Counter({0: 255560, 1: 1833})
Counter({0: 2555, 1: 2555})


CPU times: user 9min 10s, sys: 5.25 s, total: 9min 16s
Wall time: 9min 24s


%%time
basic_balancedRFC(trainX=X_SMVSMOTEIHTRUS, trainy=y_SVMSMOTEIHTRUS, testX=X_test, testy=y_test, 
                  weight='balanced')

Accuracy: 0.908
F1-binary: 0.127
Recall: 0.944


CPU times: user 1.95 s, sys: 47.6 ms, total: 1.99 s
Wall time: 1.65 s


%%time

from imblearn.pipeline import Pipeline

print(Counter(y_train))
svmsmote = SVMSMOTE(sampling_strategy=0.008, random_state=42)
iht = InstanceHardnessThreshold(random_state=42)
rus = RandomUnderSampler(random_state=42)
pipeline = Pipeline([('svmsmote', svmsmote), ('iht', iht), ('rus', rus)])
X_SMVSMOTEIHTRUS, y_SVMSMOTEIHTRUS = pipeline.fit_resample(X_train, y_train)
print(Counter(y_SVMSMOTEIHTRUS))
print('\n')

Counter({0: 255560, 1: 1833})
Counter({0: 2044, 1: 2044})


CPU times: user 8min 53s, sys: 5.13 s, total: 8min 58s
Wall time: 9min 6s


%%time
basic_balancedRFC(trainX=X_SMVSMOTEIHTRUS, trainy=y_SVMSMOTEIHTRUS, testX=X_test, testy=y_test, 
                  weight='balanced')

Accuracy: 0.897
F1-binary: 0.117
Recall: 0.952


CPU times: user 1.79 s, sys: 37 ms, total: 1.83 s
Wall time: 1.43 s


%%time

from imblearn.pipeline import Pipeline

print(Counter(y_train))
svmsmote = SVMSMOTE(sampling_strategy=0.05, random_state=42)
iht = InstanceHardnessThreshold(random_state=42)
rus = RandomUnderSampler(random_state=42)
pipeline = Pipeline([('svmsmote', svmsmote), ('iht', iht), ('rus', rus)])
X_SMVSMOTEIHTRUS, y_SVMSMOTEIHTRUS = pipeline.fit_resample(X_train, y_train)
print(Counter(y_SVMSMOTEIHTRUS))
print('\n')

Counter({0: 255560, 1: 1833})
Counter({0: 12778, 1: 12778})


CPU times: user 9min 11s, sys: 5.29 s, total: 9min 17s
Wall time: 9min 25s


%%time
basic_balancedRFC(trainX=X_SMVSMOTEIHTRUS, trainy=y_SVMSMOTEIHTRUS, testX=X_test, testy=y_test, 
                  weight='balanced')

Accuracy: 0.942
F1-binary: 0.183
Recall: 0.911


CPU times: user 3.32 s, sys: 81.7 ms, total: 3.4 s
Wall time: 3.09 s


%%time

from imblearn.pipeline import Pipeline

print(Counter(y_train))
svmsmote = SVMSMOTE(sampling_strategy=0.1)
iht = InstanceHardnessThreshold(random_state=42)
rus = RandomUnderSampler(random_state=42)
pipeline = Pipeline([('svmsmote', svmsmote), ('iht', iht), ('rus', rus)])
X_SMVSMOTEIHTRUS, y_SVMSMOTEIHTRUS = pipeline.fit_resample(X_train, y_train)
print(Counter(y_SVMSMOTEIHTRUS))
print('\n')

Counter({0: 255560, 1: 1833})
Counter({0: 25556, 1: 25556})


CPU times: user 9min 13s, sys: 5.21 s, total: 9min 18s
Wall time: 9min 25s


%%time
basic_balancedRFC(trainX=X_SMVSMOTEIHTRUS, trainy=y_SVMSMOTEIHTRUS, testX=X_test, testy=y_test, 
                  weight='balanced')

Accuracy: 0.952
F1-binary: 0.207
Recall: 0.875


CPU times: user 5.13 s, sys: 71.5 ms, total: 5.2 s
Wall time: 4.82 s


%%time

from imblearn.pipeline import Pipeline

print(Counter(y_train))
svmsmote = SVMSMOTE(sampling_strategy=0.3)
iht = InstanceHardnessThreshold(random_state=42)
rus = RandomUnderSampler(random_state=42)
pipeline = Pipeline([('svmsmote', svmsmote), ('iht', iht), ('rus', rus)])
X_SMVSMOTEIHTRUS, y_SVMSMOTEIHTRUS = pipeline.fit_resample(X_train, y_train)
print(Counter(y_SVMSMOTEIHTRUS))
print('\n')

Counter({0: 255560, 1: 1833})
Counter({0: 76668, 1: 76668})


CPU times: user 10min 19s, sys: 7.78 s, total: 10min 27s
Wall time: 1h 26min 43s


%%time
basic_balancedRFC(trainX=X_SMVSMOTEIHTRUS, trainy=y_SVMSMOTEIHTRUS, testX=X_test, testy=y_test, 
                  weight='balanced')

Accuracy: 0.961
F1-binary: 0.235
Recall: 0.835


CPU times: user 15.4 s, sys: 202 ms, total: 15.6 s
Wall time: 15.5 s


%%time

from imblearn.pipeline import Pipeline

print(Counter(y_train))
smotenc = SMOTENC(sampling_strategy=0.1, categorical_features=[0, 1,2], random_state=42)
rus = RandomUnderSampler(random_state=42)
pipeline = Pipeline([('smotenc', smotenc), ('rus', rus)])
X_SMOTENCRUS, y_SMOTENCRUS = pipeline.fit_resample(X_train, y_train)
print(Counter(y_SMOTENCRUS))
print('\n')

Counter({0: 255560, 1: 1833})
Counter({0: 25556, 1: 25556})


CPU times: user 16.3 s, sys: 486 ms, total: 16.8 s
Wall time: 17 s


%%time
basic_balancedRFC(trainX=X_SMOTENCRUS, trainy=y_SMOTENCRUS, testX=X_test, testy=y_test, 
                  weight='balanced')

Accuracy: 0.955
F1-binary: 0.211
Recall: 0.842


CPU times: user 5.37 s, sys: 137 ms, total: 5.51 s
Wall time: 5.13 s


%%time

from imblearn.pipeline import Pipeline

print(Counter(y_train))
smotenc = SMOTENC(sampling_strategy=0.008, categorical_features=[0, 1,2], random_state=1)
rus = RandomUnderSampler(random_state=1)
pipeline = Pipeline([('smotenc', smotenc), ('rus', rus)])
X_SMOTENCRUS, y_SMOTENCRUS = pipeline.fit_resample(X_train, y_train)
print(Counter(y_SMOTENCRUS))
print('\n')

Counter({0: 255560, 1: 1833})
Counter({0: 2044, 1: 2044})


CPU times: user 14.6 s, sys: 452 ms, total: 15.1 s
Wall time: 15 s


%%time
basic_balancedRFC(trainX=X_SMOTENCRUS, trainy=y_SMOTENCRUS, testX=X_test, testy=y_test, 
                  weight='balanced')

Accuracy: 0.909
F1-binary: 0.129
Recall: 0.954


CPU times: user 1.86 s, sys: 53.2 ms, total: 1.91 s
Wall time: 1.52 s


%%time

print(Counter(y_train))

smotenc = SMOTENC(sampling_strategy=0.1, categorical_features=[0, 1,2], random_state=42)
iht = InstanceHardnessThreshold(random_state=42)
pipeline = Pipeline([('smotenc', smotenc), ('iht', iht)])
X_SMOTENCIHT, y_SMOTENCIHT = pipeline.fit_resample(X_train, y_train)

print(Counter(y_SMOTENCIHT))
print('\n')

Counter({0: 255560, 1: 1833})
Counter({0: 223188, 1: 25556})


CPU times: user 2min 43s, sys: 3.16 s, total: 2min 46s
Wall time: 2min 50s


%%time
basic_balancedRFC(trainX=X_SMOTENCIHT, trainy=y_SMOTENCIHT, testX=X_test, testy=y_test, 
                  weight='balanced')

Accuracy: 0.930
F1-binary: 0.153
Recall: 0.885


CPU times: user 26.5 s, sys: 775 ms, total: 27.3 s
Wall time: 28.5 s


%%time

print(Counter(y_train))

smotenc = SMOTENC(sampling_strategy=0.1, categorical_features=[0, 1,2], random_state=42)
iht = InstanceHardnessThreshold(random_state=42)
rus = RandomUnderSampler(random_state=42)
pipeline = Pipeline([('smotenc', smotenc), ('iht', iht), ('rus', rus)])
X_SMOTENCIHTRUS, y_SMOTENCIHTRUS = pipeline.fit_resample(X_train, y_train)

print(Counter(y_SMOTENCIHTRUS))
print('\n')

Counter({0: 255560, 1: 1833})
Counter({0: 25556, 1: 25556})


CPU times: user 2min 49s, sys: 3.31 s, total: 2min 52s
Wall time: 2min 57s


%%time
basic_balancedRFC(trainX=X_SMOTENCIHTRUS, trainy=y_SMOTENCIHTRUS, testX=X_test, testy=y_test, 
                  weight='balanced')

Accuracy: 0.906
F1-binary: 0.123
Recall: 0.929


CPU times: user 4.89 s, sys: 79.3 ms, total: 4.97 s
Wall time: 4.7 s


%%time

print(Counter(y_train))

smotenc = SMOTENC(sampling_strategy=0.05, categorical_features=[0, 1,2], random_state=42)
iht = InstanceHardnessThreshold(random_state=42)
rus = RandomUnderSampler(random_state=42)
pipeline = Pipeline([('smotenc', smotenc), ('iht', iht), ('rus', rus)])
X_SMOTENCIHTRUS, y_SMOTENCIHTRUS = pipeline.fit_resample(X_train, y_train)

print(Counter(y_SMOTENCIHTRUS))
print('\n')

Counter({0: 255560, 1: 1833})
Counter({0: 12778, 1: 12778})


CPU times: user 2min 33s, sys: 2.5 s, total: 2min 36s
Wall time: 2min 39s


%%time
basic_balancedRFC(trainX=X_SMOTENCIHTRUS, trainy=y_SMOTENCIHTRUS, testX=X_test, testy=y_test, 
                  weight='balanced')

Accuracy: 0.907
F1-binary: 0.126
Recall: 0.947


CPU times: user 3.39 s, sys: 78.5 ms, total: 3.47 s
Wall time: 3.28 s


%%time

print(Counter(y_train))

smotenc = SMOTENC(sampling_strategy=0.01, categorical_features=[0, 1,2], random_state=42)
iht = InstanceHardnessThreshold(random_state=42)
rus = RandomUnderSampler(random_state=42)
pipeline = Pipeline([('smotenc', smotenc), ('iht', iht), ('rus', rus)])
X_SMOTENCIHTRUS, y_SMOTENCIHTRUS = pipeline.fit_resample(X_train, y_train)

print(Counter(y_SMOTENCIHTRUS))
print('\n')

Counter({0: 255560, 1: 1833})
Counter({0: 2555, 1: 2555})


CPU times: user 2min 18s, sys: 1.81 s, total: 2min 20s
Wall time: 2min 22s


%%time
basic_balancedRFC(trainX=X_SMOTENCIHTRUS, trainy=y_SMOTENCIHTRUS, testX=X_test, testy=y_test, 
                  weight='balanced')

Accuracy: 0.894
F1-binary: 0.115
Recall: 0.967


CPU times: user 2.01 s, sys: 56.5 ms, total: 2.06 s
Wall time: 1.78 s


%%time

print(Counter(y_train))

smotenc = SMOTENC(sampling_strategy=0.008, categorical_features=[0, 1,2], random_state=42)
iht = InstanceHardnessThreshold(random_state=42)
rus = RandomUnderSampler(random_state=42)
pipeline = Pipeline([('smotenc', smotenc), ('iht', iht), ('rus', rus)])
X_SMOTENCIHTRUS, y_SMOTENCIHTRUS = pipeline.fit_resample(X_train, y_train)

print(Counter(y_SMOTENCIHTRUS))
print('\n')

Counter({0: 255560, 1: 1833})
Counter({0: 2044, 1: 2044})


CPU times: user 2min 22s, sys: 1.92 s, total: 2min 24s
Wall time: 2min 29s


%%time
basic_balancedRFC(trainX=X_SMOTENCIHTRUS, trainy=y_SMOTENCIHTRUS, testX=X_test, testy=y_test, 
                  weight='balanced')

Accuracy: 0.899
F1-binary: 0.118
Recall: 0.947


CPU times: user 1.95 s, sys: 45 ms, total: 1.99 s
Wall time: 1.64 s


%%time

print(Counter(y_train))

iht = InstanceHardnessThreshold(random_state=42)
rus = RandomUnderSampler(random_state=42)
pipeline = Pipeline([('iht', iht), ('rus', rus)])
X_IHTRUS, y_IHTRUS = pipeline.fit_resample(X_train, y_train)

print(Counter(y_IHTRUS))
print('\n')

Counter({0: 255560, 1: 1833})
Counter({0: 1833, 1: 1833})


CPU times: user 1min 58s, sys: 1.41 s, total: 2min
Wall time: 2min 2s


%%time
basic_balancedRFC(trainX=X_IHTRUS, trainy=y_IHTRUS, testX=X_test, testy=y_test, 
                  weight='balanced')

Accuracy: 0.903
F1-binary: 0.122
Recall: 0.954


CPU times: user 1.82 s, sys: 40.7 ms, total: 1.86 s
Wall time: 1.47 s


%%time

print(Counter(y_train))

iht = InstanceHardnessThreshold(sampling_strategy=0.01, random_state=42)
rus = RandomUnderSampler(random_state=42)
pipeline = Pipeline([('iht', iht), ('rus', rus)])
X_IHTRUS, y_IHTRUS = pipeline.fit_resample(X_train, y_train)

print(Counter(y_IHTRUS))
print('\n')

Counter({0: 255560, 1: 1833})
Counter({0: 1833, 1: 1833})


CPU times: user 1min 58s, sys: 1.38 s, total: 1min 59s
Wall time: 2min


%%time
basic_balancedRFC(trainX=X_IHTRUS, trainy=y_IHTRUS, testX=X_test, testy=y_test, 
                  weight='balanced')

Accuracy: 0.903
F1-binary: 0.122
Recall: 0.954


CPU times: user 1.88 s, sys: 42.2 ms, total: 1.92 s
Wall time: 1.86 s


%%time
dill.dump_session('TFM_notebook_env.db')

CPU times: user 1min 34s, sys: 5.33 s, total: 1min 40s
Wall time: 1min 42s


%%time

print(Counter(y_train))

smotenc = SMOTENC(sampling_strategy=0.01, categorical_features=[0, 1,2], random_state=42)
iht = InstanceHardnessThreshold(random_state=42)
rus = RandomUnderSampler(random_state=42)
pipeline = Pipeline([('smotenc', smotenc), ('iht', iht), ('rus', rus)])
X_SMOTENCIHTRUS, y_SMOTENCIHTRUS = pipeline.fit_resample(X_train, y_train)

print(Counter(y_SMOTENCIHTRUS))
print('\n')
basic_balancedRFC(trainX=X_SMOTENCIHTRUS, trainy=y_SMOTENCIHTRUS, testX=X_test, testy=y_test, 
                  weight='balanced')

Counter({0: 255560, 1: 1833})

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
<timed exec> in <module>

~/opt/anaconda3/envs/TFM/lib/python3.7/site-packages/sklearn/utils/validation.py in inner_f(*args, **kwargs)
     61             extra_args = len(args) - len(all_args)
     62             if extra_args <= 0:
---> 63                 return f(*args, **kwargs)
     64 
     65             # extra_args > 0

~/opt/anaconda3/envs/TFM/lib/python3.7/site-packages/sklearn/pipeline.py in __init__(self, steps, memory, verbose)
    116         self.memory = memory
    117         self.verbose = verbose
--> 118         self._validate_steps()
    119 
    120     def get_params(self, deep=True):

~/opt/anaconda3/envs/TFM/lib/python3.7/site-packages/sklearn/pipeline.py in _validate_steps(self)
    169                                 "transformers and implement fit and transform "
    170                                 "or be the string 'passthrough' "
--> 171                                 "'%s' (type %s) doesn't" % (t, type(t)))
    172 
    173         # We allow last estimator to be None as an identity transformation

TypeError: All intermediate steps should be transformers and implement fit and transform or be the string 'passthrough' 'SMOTENC(categorical_features=[0, 1, 2], random_state=42, sampling_strategy=0.01)' (type <class 'imblearn.over_sampling._smote.SMOTENC'>) doesn't


%%time

print(Counter(y_train))

smotenc = SMOTENC(sampling_strategy=0.01, categorical_features=[0, 1,2], random_state=42)
iht = InstanceHardnessThreshold(random_state=42)
rus = RandomUnderSampler(random_state=42)
pipeline = Pipeline([('smotenc', smotenc), ('iht', iht), ('rus', rus)])
X_SMOTENCIHTRUS0, y_SMOTENCIHTRUS0 = pipeline.fit_resample(X_train0, y_train)

print(Counter(y_SMOTENCIHTRUS0))
print('\n')
basic_balancedRFC(trainX=X_SMOTENCIHTRUS0, trainy=y_SMOTENCIHTRUS0, testX=X_test0, testy=y_test, 
                  weight='balanced')

Counter({0: 255560, 1: 1833})

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
<timed exec> in <module>

~/opt/anaconda3/envs/TFM/lib/python3.7/site-packages/sklearn/utils/validation.py in inner_f(*args, **kwargs)
     61             extra_args = len(args) - len(all_args)
     62             if extra_args <= 0:
---> 63                 return f(*args, **kwargs)
     64 
     65             # extra_args > 0

~/opt/anaconda3/envs/TFM/lib/python3.7/site-packages/sklearn/pipeline.py in __init__(self, steps, memory, verbose)
    116         self.memory = memory
    117         self.verbose = verbose
--> 118         self._validate_steps()
    119 
    120     def get_params(self, deep=True):

~/opt/anaconda3/envs/TFM/lib/python3.7/site-packages/sklearn/pipeline.py in _validate_steps(self)
    169                                 "transformers and implement fit and transform "
    170                                 "or be the string 'passthrough' "
--> 171                                 "'%s' (type %s) doesn't" % (t, type(t)))
    172 
    173         # We allow last estimator to be None as an identity transformation

TypeError: All intermediate steps should be transformers and implement fit and transform or be the string 'passthrough' 'SMOTENC(categorical_features=[0, 1, 2], random_state=42, sampling_strategy=0.01)' (type <class 'imblearn.over_sampling._smote.SMOTENC'>) doesn't


%%time

print(Counter(y_train))

smotenc = SMOTENC(sampling_strategy=0.01, categorical_features=[0, 1,2], random_state=42)
iht = InstanceHardnessThreshold(random_state=42)
rus = RandomUnderSampler(sampling_strategy=0.9, random_state=42)
pipeline = Pipeline([('smotenc', smotenc), ('iht', iht), ('rus', rus)])
X_SMOTENCIHTRUS1, y_SMOTENCIHTRUS1 = pipeline.fit_resample(X_train, y_train)

print(Counter(y_SMOTENCIHTRUS1))
print('\n')
basic_balancedRFC(trainX=X_SMOTENCIHTRUS1, trainy=y_SMOTENCIHTRUS1, testX=X_test, testy=y_test, 
                  weight='balanced')

Counter({0: 255560, 1: 1833})
Counter({0: 2838, 1: 2555})


Accuracy: 0.899
F1-binary: 0.120
Recall: 0.962


CPU times: user 2min 32s, sys: 2.39 s, total: 2min 35s
Wall time: 2min 42s


%%time

print(Counter(y_train))

smotenc = SMOTENC(sampling_strategy=0.01, categorical_features=[0, 1,2], random_state=42)
iht = InstanceHardnessThreshold(random_state=42)
rus = RandomUnderSampler(sampling_strategy=0.8, random_state=42)
pipeline = Pipeline([('smotenc', smotenc), ('iht', iht), ('rus', rus)])
X_SMOTENCIHTRUS2, y_SMOTENCIHTRUS2 = pipeline.fit_resample(X_train, y_train)

print(Counter(y_SMOTENCIHTRUS2))
print('\n')
basic_balancedRFC(trainX=X_SMOTENCIHTRUS2, trainy=y_SMOTENCIHTRUS2, testX=X_test, testy=y_test, 
                  weight='balanced')

Counter({0: 255560, 1: 1833})
Counter({0: 3193, 1: 2555})


Accuracy: 0.902
F1-binary: 0.123
Recall: 0.959


CPU times: user 2min 22s, sys: 1.94 s, total: 2min 24s
Wall time: 2min 26s


%%time

print(Counter(y_train))

smotenc = SMOTENC(sampling_strategy=0.01, categorical_features=[0, 1,2], random_state=42)
iht = InstanceHardnessThreshold(random_state=42)
rus = RandomUnderSampler(sampling_strategy=0.7, random_state=42)
pipeline = Pipeline([('smotenc', smotenc), ('iht', iht), ('rus', rus)])
X_SMOTENCIHTRUS3, y_SMOTENCIHTRUS3 = pipeline.fit_resample(X_train, y_train)

print(Counter(y_SMOTENCIHTRUS3))
print('\n')
basic_balancedRFC(trainX=X_SMOTENCIHTRUS3, trainy=y_SMOTENCIHTRUS3, testX=X_test, testy=y_test, 
                  weight='balanced')

Counter({0: 255560, 1: 1833})
Counter({0: 3650, 1: 2555})


Accuracy: 0.908
F1-binary: 0.129
Recall: 0.954


CPU times: user 2min 23s, sys: 1.89 s, total: 2min 25s
Wall time: 2min 28s


dill.dump_session('TFM_notebook_env.db')


#import dill
#dill.load_session('TFM_notebook_env.db')


%%time
basic_balancedLRC(trainX=X_SMOTENCIHTRUS, trainy=y_SMOTENCIHTRUS, testX=X_test, testy=y_test, 
                  weight='balanced')

/Users/inigoab/opt/anaconda3/envs/TFM/lib/python3.7/site-packages/sklearn/linear_model/_logistic.py:765: ConvergenceWarning: lbfgs failed to converge (status=1):
STOP: TOTAL NO. of ITERATIONS REACHED LIMIT.

Increase the number of iterations (max_iter) or scale the data as shown in:
    https://scikit-learn.org/stable/modules/preprocessing.html
Please also refer to the documentation for alternative solver options:
    https://scikit-learn.org/stable/modules/linear_model.html#logistic-regression
  extra_warning_msg=_LOGISTIC_SOLVER_CONVERGENCE_MSG)

Accuracy: 0.902
F1-binary: 0.117
Recall: 0.911


CPU times: user 1.4 s, sys: 190 ms, total: 1.59 s
Wall time: 705 ms


%%time

balance = [{0:1,1:1}, {0:1,1:2}, {0:1,1:3}]

grid_LRC(trainX=X_SMOTENCIHTRUS, trainy=y_SMOTENCIHTRUS, testX=X_test, testy=y_test, 
                  balance=balance)

/Users/inigoab/opt/anaconda3/envs/TFM/lib/python3.7/site-packages/sklearn/linear_model/_logistic.py:765: ConvergenceWarning: lbfgs failed to converge (status=1):
STOP: TOTAL NO. of ITERATIONS REACHED LIMIT.

Increase the number of iterations (max_iter) or scale the data as shown in:
    https://scikit-learn.org/stable/modules/preprocessing.html
Please also refer to the documentation for alternative solver options:
    https://scikit-learn.org/stable/modules/linear_model.html#logistic-regression
  extra_warning_msg=_LOGISTIC_SOLVER_CONVERGENCE_MSG)
[Parallel(n_jobs=1)]: Using backend SequentialBackend with 1 concurrent workers.
/Users/inigoab/opt/anaconda3/envs/TFM/lib/python3.7/site-packages/sklearn/linear_model/_logistic.py:765: ConvergenceWarning: lbfgs failed to converge (status=1):
STOP: TOTAL NO. of ITERATIONS REACHED LIMIT.

Increase the number of iterations (max_iter) or scale the data as shown in:
    https://scikit-learn.org/stable/modules/preprocessing.html
Please also refer to the documentation for alternative solver options:
    https://scikit-learn.org/stable/modules/linear_model.html#logistic-regression
  extra_warning_msg=_LOGISTIC_SOLVER_CONVERGENCE_MSG)
[Parallel(n_jobs=1)]: Done   1 out of   1 | elapsed:    0.1s finished

El mejor parámetros es: {'class_weight': {0: 1, 1: 3}}
Accuracy: 0.843
F1-binary: 0.080
Recall: 0.959


CPU times: user 2.33 s, sys: 445 ms, total: 2.77 s
Wall time: 9.65 s


%%time
basic_GNB(trainX=X_SMOTENCIHTRUS, trainy=y_SMOTENCIHTRUS, testX=X_test, testy=y_test)

Accuracy: 0.823
F1-binary: 0.060
Recall: 0.789


CPU times: user 750 ms, sys: 45.8 ms, total: 796 ms
Wall time: 508 ms


%%time
basic_KNN(trainX=X_SMOTENCIHTRUS, trainy=y_SMOTENCIHTRUS, testX=X_test, testy=y_test)

Accuracy: 0.890
F1-binary: 0.105
Recall: 0.906


CPU times: user 6.01 s, sys: 119 ms, total: 6.13 s
Wall time: 6.01 s


%%time
basic_balancedDTC(trainX=X_SMOTENCIHTRUS, trainy=y_SMOTENCIHTRUS, testX=X_test, testy=y_test, 
                  weight='balanced')

Accuracy: 0.900
F1-binary: 0.119
Recall: 0.944


CPU times: user 694 ms, sys: 33.3 ms, total: 727 ms
Wall time: 516 ms


%%time
basic_balancedDTC(trainX=X_SMOTENCIHTRUS, trainy=y_SMOTENCIHTRUS, testX=X_test, testy=y_test, 
                  weight=50)

Accuracy: 0.900
F1-binary: 0.119
Recall: 0.944


CPU times: user 759 ms, sys: 36.3 ms, total: 796 ms
Wall time: 498 ms


%%time
balancedXGB(trainX=X_SMOTENCIHTRUS, trainy=y_SMOTENCIHTRUS, testX=X_test, testy=y_test, 
            weight='balanced')

Peso estimado:  1.0
Accuracy: 0.907
F1-binary: 0.129
Recall: 0.969

CPU times: user 2.86 s, sys: 292 ms, total: 3.15 s
Wall time: 1.61 s


%%time
balancedXGB(trainX=X_SMOTENCIHTRUS, trainy=y_SMOTENCIHTRUS, testX=X_test, testy=y_test, 
            weight=25)

Accuracy: 0.871
F1-binary: 0.098
Recall: 0.982

CPU times: user 2.88 s, sys: 233 ms, total: 3.11 s
Wall time: 1.55 s


%%time
basic_balancedRFC(trainX=X_SMOTENCIHTRUS, trainy=y_SMOTENCIHTRUS, testX=X_test, testy=y_test, 
                  weight='balanced_subsample')

Accuracy: 0.894
F1-binary: 0.115
Recall: 0.964


CPU times: user 2.35 s, sys: 62.5 ms, total: 2.41 s
Wall time: 2.12 s


%%time
basic_balancedRFC(trainX=X_SMOTENCIHTRUS, trainy=y_SMOTENCIHTRUS, testX=X_test, testy=y_test, 
                  weight='balanced')

Accuracy: 0.894
F1-binary: 0.115
Recall: 0.967


CPU times: user 2.15 s, sys: 86.9 ms, total: 2.24 s
Wall time: 2.01 s


%%time
basic_balancedRFC(trainX=X_SMOTENCIHTRUS, trainy=y_SMOTENCIHTRUS, testX=X_test, testy=y_test, 
                  weight={0:1,1:5})

Accuracy: 0.897
F1-binary: 0.119
Recall: 0.969


CPU times: user 2.02 s, sys: 52.3 ms, total: 2.07 s
Wall time: 1.75 s


dill.dump_session('TFM_notebook_env.db')


%%time
balancedRFC(trainX=X_SMOTENCIHTRUS, trainy=y_SMOTENCIHTRUS,
            testX=X_test, testy=y_test, 
            scoring='recall', weight='balanced',
            n_estimators=[50, 100, 200, 250, 300, 350, 400], 
            max_depth=range(5,25),
            standardize=False, normalize=False, pca=False)

Fitting 20 folds for each of 140 candidates, totalling 2800 fits

Los mejores parámetros son: {'rfc__max_depth': 11, 'rfc__n_estimators': 200}

[Parallel(n_jobs=1)]: Using backend SequentialBackend with 1 concurrent workers.
[Parallel(n_jobs=1)]: Done 200 out of 200 | elapsed:    0.8s finished
[Parallel(n_jobs=1)]: Using backend SequentialBackend with 1 concurrent workers.
[Parallel(n_jobs=1)]: Done 200 out of 200 | elapsed:    1.4s finished

Accuracy: 0.901
F1-binary: 0.122
Recall: 0.967


CPU times: user 36.1 s, sys: 4.1 s, total: 40.2 s
Wall time: 33min 45s


%%time
balancedRFC(trainX=X_SMOTENCIHTRUS, trainy=y_SMOTENCIHTRUS,
            testX=X_test, testy=y_test, 
            scoring='recall', weight='balanced',
            n_estimators=[50, 100, 200, 250], 
            max_depth=range(5,12),
            standardize=True, normalize=True, pca=False)

Fitting 20 folds for each of 28 candidates, totalling 560 fits

Los mejores parámetros son: {'rfc__max_depth': 11, 'rfc__n_estimators': 100}

[Parallel(n_jobs=1)]: Using backend SequentialBackend with 1 concurrent workers.
[Parallel(n_jobs=1)]: Done 100 out of 100 | elapsed:    0.5s finished
[Parallel(n_jobs=1)]: Using backend SequentialBackend with 1 concurrent workers.
[Parallel(n_jobs=1)]: Done 100 out of 100 | elapsed:    0.7s finished

Accuracy: 0.902
F1-binary: 0.122
Recall: 0.964


CPU times: user 10.7 s, sys: 988 ms, total: 11.6 s
Wall time: 4min


%%time
balancedRFC(trainX=X_SMOTENCIHTRUS, trainy=y_SMOTENCIHTRUS,
            testX=X_test, testy=y_test, 
            scoring='recall', weight='balanced',
            n_estimators=[50, 100, 200, 250], 
            max_depth=range(5,12),
            standardize=False, normalize=False, pca=False)

Fitting 20 folds for each of 28 candidates, totalling 560 fits

Los mejores parámetros son: {'rfc__max_depth': 11, 'rfc__n_estimators': 200}

[Parallel(n_jobs=1)]: Using backend SequentialBackend with 1 concurrent workers.
[Parallel(n_jobs=1)]: Done 200 out of 200 | elapsed:    1.1s finished
[Parallel(n_jobs=1)]: Using backend SequentialBackend with 1 concurrent workers.
[Parallel(n_jobs=1)]: Done 200 out of 200 | elapsed:    1.6s finished

Accuracy: 0.901
F1-binary: 0.122
Recall: 0.967


CPU times: user 12.3 s, sys: 1.02 s, total: 13.3 s
Wall time: 3min 52s


%%time
balancedRFC(trainX=X_SMOTENCIHTRUS, trainy=y_SMOTENCIHTRUS,
            testX=X_test, testy=y_test, 
            scoring='recall', weight='balanced',
            n_estimators=[100,150,200], 
            max_depth=range(10,12),
            standardize=True, normalize=True, pca=True)

Fitting 20 folds for each of 6 candidates, totalling 120 fits

Los mejores parámetros son: {'rfc__max_depth': 11, 'rfc__n_estimators': 200}

[Parallel(n_jobs=1)]: Using backend SequentialBackend with 1 concurrent workers.
[Parallel(n_jobs=1)]: Done 200 out of 200 | elapsed:    1.0s finished
[Parallel(n_jobs=1)]: Using backend SequentialBackend with 1 concurrent workers.
[Parallel(n_jobs=1)]: Done 200 out of 200 | elapsed:    1.3s finished

Accuracy: 0.894
F1-binary: 0.116
Recall: 0.972


CPU times: user 7.52 s, sys: 374 ms, total: 7.9 s
Wall time: 2min 2s


%%time
balancedRFC(trainX=X_SMOTENCIHTRUS, trainy=y_SMOTENCIHTRUS,
            testX=X_test, testy=y_test, 
            scoring='recall', weight='balanced',
            n_estimators=[50, 75], 
            max_depth=range(10,11),
            standardize=False, normalize=False, pca=False)

Fitting 20 folds for each of 2 candidates, totalling 40 fits

Los mejores parámetros son: {'rfc__max_depth': 10, 'rfc__n_estimators': 75}

[Parallel(n_jobs=1)]: Using backend SequentialBackend with 1 concurrent workers.
[Parallel(n_jobs=1)]: Done  75 out of  75 | elapsed:    0.4s finished
[Parallel(n_jobs=1)]: Using backend SequentialBackend with 1 concurrent workers.
[Parallel(n_jobs=1)]: Done  75 out of  75 | elapsed:    0.5s finished

Accuracy: 0.893
F1-binary: 0.115
Recall: 0.975


CPU times: user 2.46 s, sys: 109 ms, total: 2.57 s
Wall time: 7.83 s


%%time
balancedRFC(trainX=X_SMOTENCIHTRUS, trainy=y_SMOTENCIHTRUS,
            testX=X_test, testy=y_test, 
            scoring='recall', weight='balanced',
            n_estimators=range(60,90), 
            max_depth=[10],
            standardize=False, normalize=False, pca=False)

Fitting 20 folds for each of 30 candidates, totalling 600 fits

Los mejores parámetros son: {'rfc__max_depth': 10, 'rfc__n_estimators': 80}

[Parallel(n_jobs=1)]: Using backend SequentialBackend with 1 concurrent workers.
[Parallel(n_jobs=1)]: Done  80 out of  80 | elapsed:    0.4s finished
[Parallel(n_jobs=1)]: Using backend SequentialBackend with 1 concurrent workers.
[Parallel(n_jobs=1)]: Done  80 out of  80 | elapsed:    0.6s finished

Accuracy: 0.893
F1-binary: 0.115
Recall: 0.975


CPU times: user 9.79 s, sys: 986 ms, total: 10.8 s
Wall time: 2min 30s


%%time
balancedRFC(trainX=X_SMOTENCIHTRUS, trainy=y_SMOTENCIHTRUS,
            testX=X_test, testy=y_test, 
            scoring='recall', 
            #weight='balanced',
            n_estimators=[75], 
            max_depth=[10],
            standardize=False, normalize=False, pca=False)

Fitting 20 folds for each of 1 candidates, totalling 20 fits

Los mejores parámetros son: {'rfc__max_depth': 10, 'rfc__n_estimators': 75}

[Parallel(n_jobs=1)]: Using backend SequentialBackend with 1 concurrent workers.
[Parallel(n_jobs=1)]: Done  75 out of  75 | elapsed:    0.3s finished
[Parallel(n_jobs=1)]: Using backend SequentialBackend with 1 concurrent workers.
[Parallel(n_jobs=1)]: Done  75 out of  75 | elapsed:    0.5s finished

Accuracy: 0.893
F1-binary: 0.115
Recall: 0.975


CPU times: user 2.19 s, sys: 83.6 ms, total: 2.28 s
Wall time: 6.01 s


%%time

balance =[{0:1,1:3.5}, {0:1,1:5}, {0:1,1:8}]

grid_balancedRFC(trainX=X_SMOTENCIHTRUS, trainy=y_SMOTENCIHTRUS,
                 testX=X_test, testy=y_test, 
                scoring='recall', weights=balance,
                 n_estimators=[75], 
                 max_depth=[10],
                standardize=False, normalize=False, pca=False)

Fitting 20 folds for each of 3 candidates, totalling 60 fits

Los mejores parámetros son: {'rfc__class_weight': {0: 1, 1: 8}, 'rfc__max_depth': 10, 'rfc__n_estimators': 75}

[Parallel(n_jobs=1)]: Using backend SequentialBackend with 1 concurrent workers.
[Parallel(n_jobs=1)]: Done  75 out of  75 | elapsed:    0.4s finished
[Parallel(n_jobs=1)]: Using backend SequentialBackend with 1 concurrent workers.
[Parallel(n_jobs=1)]: Done  75 out of  75 | elapsed:    0.5s finished

Accuracy: 0.859
F1-binary: 0.090
Recall: 0.975


CPU times: user 2.73 s, sys: 131 ms, total: 2.86 s
Wall time: 13.9 s


%%time

balance =[{0:1,1:3.5}]

grid_balancedRFC(trainX=X_SMOTENCIHTRUS, trainy=y_SMOTENCIHTRUS,
                 testX=X_test, testy=y_test, 
                scoring='recall', weights=balance,
                 n_estimators=[50, 75, 100], max_depth=range(5,12),
                standardize=False, normalize=False, pca=False)

Fitting 20 folds for each of 21 candidates, totalling 420 fits

Los mejores parámetros son: {'rfc__class_weight': {0: 1, 1: 3.5}, 'rfc__max_depth': 8, 'rfc__n_estimators': 50}

[Parallel(n_jobs=1)]: Using backend SequentialBackend with 1 concurrent workers.
[Parallel(n_jobs=1)]: Done  50 out of  50 | elapsed:    0.3s finished
[Parallel(n_jobs=1)]: Using backend SequentialBackend with 1 concurrent workers.
[Parallel(n_jobs=1)]: Done  50 out of  50 | elapsed:    0.9s finished

Accuracy: 0.861
F1-binary: 0.091
Recall: 0.975


CPU times: user 7.33 s, sys: 629 ms, total: 7.96 s
Wall time: 3min 3s


%%time

grid_balancedDTC(trainX=X_SMOTENCIHTRUS, trainy=y_SMOTENCIHTRUS, testX=X_test, testy=y_test, 
                 max_depth= range(2, 50),
                 min_samples_split= [2, 3, 5],
                 class_weight= [{0:1,1:1.02},{0:1,1:1.05}],
                 scoring= 'recall')

Fitting 30 folds for each of 288 candidates, totalling 8640 fits

Los mejores parámetros son: {'class_weight': {0: 1, 1: 1.02}, 'max_depth': 4, 'min_samples_split': 2}
Accuracy: 0.858
F1-binary: 0.089
Recall: 0.969


CPU times: user 42 s, sys: 5 s, total: 47 s
Wall time: 1min 54s


%%time

grid_balancedDTC(trainX=X_SMOTENCIHTRUS, trainy=y_SMOTENCIHTRUS, testX=X_test, testy=y_test, 
                 max_depth= [4],
                 min_samples_split= [2],
                 #class_weight= [{0:1,1:1}],
                 scoring= 'recall')

Fitting 30 folds for each of 3 candidates, totalling 90 fits

Los mejores parámetros son: {'class_weight': {0: 1, 1: 100}, 'max_depth': 4, 'min_samples_split': 2}
Accuracy: 0.858
F1-binary: 0.089
Recall: 0.969


CPU times: user 1.15 s, sys: 112 ms, total: 1.26 s
Wall time: 1.86 s


%%time

grid_balancedDTC(trainX=X_SMOTENCIHTRUS, trainy=y_SMOTENCIHTRUS, testX=X_test, testy=y_test, 
                 max_depth= range(2, 6),
                 min_samples_split= [2, 4],
                 class_weight= [
                     #{0:1,1:1.01}, 
                     #{0:1,1:1.02},
                     #{0:1,1:1.05},
                     {0:1,1:1},
                               ],
                 scoring= 'recall')

Fitting 30 folds for each of 8 candidates, totalling 240 fits

Los mejores parámetros son: {'class_weight': {0: 1, 1: 1}, 'max_depth': 4, 'min_samples_split': 2}
Accuracy: 0.858
F1-binary: 0.089
Recall: 0.969


CPU times: user 1.96 s, sys: 162 ms, total: 2.12 s
Wall time: 2.69 s


#import dill
#dill.load_session('TFM_notebook_env.db')


%%time

rfc= RandomForestClassifier(random_state=42,
                            n_estimators=25,
                            max_depth=5,
                            class_weight= 'balanced', 
                            bootstrap=False)

rfc.fit(X_SMOTENCIHTRUS, y_SMOTENCIHTRUS)
    
pred = rfc.predict(X_val)
print("Accuracy: {:.3f}".format(accuracy_score(y_val, pred)))
print("F1-binary: {:.3f}".format(f1_score(y_val, pred, average='binary'))) 
print("Recall: {:.3f}".format(recall_score(y_val, pred))) 

matriz = confusion_matrix(y_val, pred)
sns.heatmap(matriz, annot=True, fmt="d", cmap='Blues', square=True)
plt.xlabel("predicción")
plt.ylabel("real")
plt.title('Matriz de confusión', y=1.05, size=18)
plt.show()
print('\n')

Accuracy: 0.868
F1-binary: 0.094
Recall: 0.964


CPU times: user 844 ms, sys: 37.6 ms, total: 882 ms
Wall time: 1.03 s


%%time

from sklearn.tree import plot_tree

estimator = 0

plt.subplots(figsize=(55,25))
plot_tree(rfc.estimators_[estimator],
          feature_names = X_train.columns, 
          class_names = ['Usuario', 'Cliente'],
          filled=True, node_ids=True, fontsize=10)

plt.title(f'Estimator: {estimator}', fontsize = 20)

plt.savefig(f'rf_tree_{estimator}.png')

plt.show()

CPU times: user 13.3 s, sys: 716 ms, total: 14 s
Wall time: 19.7 s


%%time

from sklearn.tree import plot_tree

estimator = 1

plt.subplots(figsize=(55,25))
plot_tree(rfc.estimators_[estimator],
          feature_names = X_train.columns, 
          class_names = ['Usuario', 'Cliente'],
          filled=True, node_ids=True, fontsize=10)

plt.title(f'Estimator: {estimator}', fontsize = 20)

plt.savefig(f'rf_tree_{estimator}.png')
    
plt.show()

CPU times: user 11.5 s, sys: 601 ms, total: 12.1 s
Wall time: 17.7 s


%%time

from sklearn.tree import plot_tree

estimator = 2

plt.subplots(figsize=(55,25))
plot_tree(rfc.estimators_[estimator],
          feature_names = X_train.columns, 
          class_names = ['Usuario', 'Cliente'],
          filled=True, node_ids=True, fontsize=10)

plt.title(f'Estimator: {estimator}', fontsize = 20)

plt.savefig(f'rf_tree_{estimator}.png')
    
plt.show()

CPU times: user 12.9 s, sys: 642 ms, total: 13.5 s
Wall time: 22.2 s


%%time

from sklearn.tree import plot_tree

estimator = 3

plt.subplots(figsize=(55,25))
plot_tree(rfc.estimators_[estimator],
          feature_names = X_train.columns, 
          class_names = ['Usuario', 'Cliente'],
          filled=True, node_ids=True, fontsize=10)

plt.title(f'Estimator: {estimator}', fontsize = 20)

plt.savefig(f'rf_tree_{estimator}.png')
    
plt.show()

CPU times: user 12.2 s, sys: 654 ms, total: 12.9 s
Wall time: 18.3 s


%%time

from sklearn.tree import plot_tree

estimator = 4

plt.subplots(figsize=(55,25))
plot_tree(rfc.estimators_[estimator],
          feature_names = X_train.columns, 
          class_names = ['Usuario', 'Cliente'],
          filled=True, node_ids=True, fontsize=10)

plt.title(f'Estimator: {estimator}', fontsize = 20)

plt.savefig(f'rf_tree_{estimator}.png')
    
plt.show()

CPU times: user 11.8 s, sys: 648 ms, total: 12.5 s
Wall time: 15.8 s


%%time

from sklearn.tree import plot_tree

estimator = 5

plt.subplots(figsize=(55,25))
plot_tree(rfc.estimators_[estimator],
          feature_names = X_train.columns, 
          class_names = ['Usuario', 'Cliente'],
          filled=True, node_ids=True, fontsize=10)

plt.title(f'Estimator: {estimator}', fontsize = 20)

plt.savefig(f'rf_tree_{estimator}.png')
    
plt.show()

CPU times: user 12.5 s, sys: 675 ms, total: 13.1 s
Wall time: 19 s


%%time

from sklearn.tree import plot_tree

estimator = 6

plt.subplots(figsize=(55,25))
plot_tree(rfc.estimators_[estimator],
          feature_names = X_train.columns, 
          class_names = ['Usuario', 'Cliente'],
          filled=True, node_ids=True, fontsize=10)

plt.title(f'Estimator: {estimator}', fontsize = 20)

plt.savefig(f'rf_tree_{estimator}.png')
    
plt.show()

CPU times: user 14.1 s, sys: 817 ms, total: 14.9 s
Wall time: 23.3 s


%%time

from sklearn.tree import plot_tree

estimator = 7

plt.subplots(figsize=(55,25))
plot_tree(rfc.estimators_[estimator],
          feature_names = X_train.columns, 
          class_names = ['Usuario', 'Cliente'],
          filled=True, node_ids=True, fontsize=10)

plt.title(f'Estimator: {estimator}', fontsize = 20)

plt.savefig(f'rf_tree_{estimator}.png')
    
plt.show()

CPU times: user 9.82 s, sys: 453 ms, total: 10.3 s
Wall time: 14.5 s


%%time

from sklearn.tree import plot_tree

estimator = 8

plt.subplots(figsize=(55,25))
plot_tree(rfc.estimators_[estimator],
          feature_names = X_train.columns, 
          class_names = ['Usuario', 'Cliente'],
          filled=True, node_ids=True, fontsize=10)

plt.title(f'Estimator: {estimator}', fontsize = 20)

plt.savefig(f'rf_tree_{estimator}.png')
    
plt.show()

CPU times: user 10.5 s, sys: 565 ms, total: 11.1 s
Wall time: 13.1 s


%%time

from sklearn.tree import plot_tree

estimator = 9

plt.subplots(figsize=(55,25))
plot_tree(rfc.estimators_[estimator],
          feature_names = X_train.columns, 
          class_names = ['Usuario', 'Cliente'],
          filled=True, node_ids=True, fontsize=10)

plt.title(f'Estimator: {estimator}', fontsize = 20)

plt.savefig(f'rf_tree_{estimator}.png')
    
plt.show()

CPU times: user 12.5 s, sys: 602 ms, total: 13.1 s
Wall time: 18 s


%%time

from sklearn.tree import plot_tree

estimator = 10

plt.subplots(figsize=(55,25))
plot_tree(rfc.estimators_[estimator],
          feature_names = X_train.columns, 
          class_names = ['Usuario', 'Cliente'],
          filled=True, node_ids=True, fontsize=10)

plt.title(f'Estimator: {estimator}', fontsize = 20)

plt.savefig(f'rf_tree_{estimator}.png')
    
plt.show()

CPU times: user 12.1 s, sys: 568 ms, total: 12.6 s
Wall time: 19.9 s


%%time

from sklearn.tree import plot_tree

estimator = 11

plt.subplots(figsize=(55,25))
plot_tree(rfc.estimators_[estimator],
          feature_names = X_train.columns, 
          class_names = ['Usuario', 'Cliente'],
          filled=True, node_ids=True, fontsize=10)

plt.title(f'Estimator: {estimator}', fontsize = 20)

plt.savefig(f'rf_tree_{estimator}.png')
    
plt.show()

CPU times: user 11.2 s, sys: 564 ms, total: 11.7 s
Wall time: 15.6 s


%%time

from sklearn.tree import plot_tree

estimator = 12

plt.subplots(figsize=(55,25))
plot_tree(rfc.estimators_[estimator],
          feature_names = X_train.columns, 
          class_names = ['Usuario', 'Cliente'],
          filled=True, node_ids=True, fontsize=10)

plt.title(f'Estimator: {estimator}', fontsize = 20)

plt.savefig(f'rf_tree_{estimator}.png')
    
plt.show()

CPU times: user 12.4 s, sys: 650 ms, total: 13 s
Wall time: 17.8 s


%%time

from sklearn.tree import plot_tree

estimator = 13

plt.subplots(figsize=(55,25))
plot_tree(rfc.estimators_[estimator],
          feature_names = X_train.columns, 
          class_names = ['Usuario', 'Cliente'],
          filled=True, node_ids=True, fontsize=10)

plt.title(f'Estimator: {estimator}', fontsize = 20)

plt.savefig(f'rf_tree_{estimator}.png')
    
plt.show()

CPU times: user 11.9 s, sys: 621 ms, total: 12.5 s
Wall time: 16.3 s


%%time

from sklearn.tree import plot_tree

estimator = 14

plt.subplots(figsize=(55,25))
plot_tree(rfc.estimators_[estimator],
          feature_names = X_train.columns, 
          class_names = ['Usuario', 'Cliente'],
          filled=True, node_ids=True, fontsize=10)

plt.title(f'Estimator: {estimator}', fontsize = 20)

plt.savefig(f'rf_tree_{estimator}.png')
    
plt.show()

CPU times: user 10.9 s, sys: 628 ms, total: 11.5 s
Wall time: 12.1 s


%%time

from sklearn.tree import plot_tree

estimator = 15

plt.subplots(figsize=(55,25))
plot_tree(rfc.estimators_[estimator],
          feature_names = X_train.columns, 
          class_names = ['Usuario', 'Cliente'],
          filled=True, node_ids=True, fontsize=10)

plt.title(f'Estimator: {estimator}', fontsize = 20)

plt.savefig(f'rf_tree_{estimator}.png')
    
plt.show()

CPU times: user 11.2 s, sys: 595 ms, total: 11.8 s
Wall time: 15.4 s


%%time

from sklearn.tree import plot_tree

estimator = 16

plt.subplots(figsize=(55,25))
plot_tree(rfc.estimators_[estimator],
          feature_names = X_train.columns, 
          class_names = ['Usuario', 'Cliente'],
          filled=True, node_ids=True, fontsize=10)

plt.title(f'Estimator: {estimator}', fontsize = 20)

plt.savefig(f'rf_tree_{estimator}.png')
    
plt.show()

CPU times: user 12 s, sys: 629 ms, total: 12.6 s
Wall time: 17.3 s


%%time

from sklearn.tree import plot_tree

estimator = 17

plt.subplots(figsize=(55,25))
plot_tree(rfc.estimators_[estimator],
          feature_names = X_train.columns, 
          class_names = ['Usuario', 'Cliente'],
          filled=True, node_ids=True, fontsize=10)

plt.title(f'Estimator: {estimator}', fontsize = 20)

plt.savefig(f'rf_tree_{estimator}.png')
    
plt.show()

CPU times: user 12.2 s, sys: 597 ms, total: 12.8 s
Wall time: 18.9 s


%%time

from sklearn.tree import plot_tree

estimator = 18

plt.subplots(figsize=(55,25))
plot_tree(rfc.estimators_[estimator],
          feature_names = X_train.columns, 
          class_names = ['Usuario', 'Cliente'],
          filled=True, node_ids=True, fontsize=10)

plt.title(f'Estimator: {estimator}', fontsize = 20)

plt.savefig(f'rf_tree_{estimator}.png')
    
plt.show()

CPU times: user 13.6 s, sys: 732 ms, total: 14.3 s
Wall time: 20.4 s


%%time

from sklearn.tree import plot_tree

estimator = 19

plt.subplots(figsize=(55,25))
plot_tree(rfc.estimators_[estimator],
          feature_names = X_train.columns, 
          class_names = ['Usuario', 'Cliente'],
          filled=True, node_ids=True, fontsize=10)

plt.title(f'Estimator: {estimator}', fontsize = 20)

plt.savefig(f'rf_tree_{estimator}.png')
    
plt.show()

CPU times: user 14.4 s, sys: 731 ms, total: 15.1 s
Wall time: 20.7 s


%%time

from sklearn.tree import plot_tree

estimator = 20

plt.subplots(figsize=(55,25))
plot_tree(rfc.estimators_[estimator],
          feature_names = X_train.columns, 
          class_names = ['Usuario', 'Cliente'],
          filled=True, node_ids=True, fontsize=10)

plt.title(f'Estimator: {estimator}', fontsize = 20)

plt.savefig(f'rf_tree_{estimator}.png')
    
plt.show()

CPU times: user 11.3 s, sys: 465 ms, total: 11.8 s
Wall time: 19.8 s


%%time

from sklearn.tree import plot_tree

estimator = 21

plt.subplots(figsize=(55,25))
plot_tree(rfc.estimators_[estimator],
          feature_names = X_train.columns, 
          class_names = ['Usuario', 'Cliente'],
          filled=True, node_ids=True, fontsize=10)

plt.title(f'Estimator: {estimator}', fontsize = 20)

plt.savefig(f'rf_tree_{estimator}.png')
    
plt.show()

CPU times: user 12.6 s, sys: 658 ms, total: 13.2 s
Wall time: 23.8 s


%%time

from sklearn.tree import plot_tree

estimator = 22

plt.subplots(figsize=(55,25))
plot_tree(rfc.estimators_[estimator],
          feature_names = X_train.columns, 
          class_names = ['Usuario', 'Cliente'],
          filled=True, node_ids=True, fontsize=10)

plt.title(f'Estimator: {estimator}', fontsize = 20)

plt.savefig(f'rf_tree_{estimator}.png')
    
plt.show()

CPU times: user 12.5 s, sys: 641 ms, total: 13.1 s
Wall time: 20.2 s


%%time

from sklearn.tree import plot_tree

estimator = 23

plt.subplots(figsize=(55,25))
plot_tree(rfc.estimators_[estimator],
          feature_names = X_train.columns, 
          class_names = ['Usuario', 'Cliente'],
          filled=True, node_ids=True, fontsize=10)

plt.title(f'Estimator: {estimator}', fontsize = 20)

plt.savefig(f'rf_tree_{estimator}.png')
    
plt.show()

CPU times: user 10.5 s, sys: 520 ms, total: 11 s
Wall time: 14.1 s


%%time

from sklearn.tree import plot_tree

estimator = 24

plt.subplots(figsize=(55,25))
plot_tree(rfc.estimators_[estimator],
          feature_names = X_train.columns, 
          class_names = ['Usuario', 'Cliente'],
          filled=True, node_ids=True, fontsize=10)

plt.title(f'Estimator: {estimator}', fontsize = 20)

plt.savefig(f'rf_tree_{estimator}.png')
    
plt.show()

CPU times: user 13.2 s, sys: 663 ms, total: 13.9 s
Wall time: 17.8 s


%%time

fig, axes = plt.subplots(nrows = 3,ncols = 1,figsize = (50,80))

for index in range(0, 3):
    tree.plot_tree(rfc.estimators_[index],
                   feature_names = X_train.columns, 
                   class_names=['Usuario', 'Cliente'],
                   filled = True, node_ids=True,
                   ax = axes[index]);
    axes[index].set_title('Estimator: ' + str(index), fontsize = 11)
    
    
plt.show()

CPU times: user 22.6 s, sys: 1.34 s, total: 23.9 s
Wall time: 25.6 s


%%time

rfc= RandomForestClassifier(random_state=42,
                            n_estimators=25,
                            max_depth=5,
                            class_weight= {0:1,1:3})

rfc.fit(X_SMOTENCIHTRUS, y_SMOTENCIHTRUS)
    
pred = rfc.predict(X_val)
print("Accuracy: {:.3f}".format(accuracy_score(y_val, pred)))
print("F1-binary: {:.3f}".format(f1_score(y_val, pred, average='binary'))) 
print("Recall: {:.3f}".format(recall_score(y_val, pred))) 

matriz = confusion_matrix(y_val, pred)
sns.heatmap(matriz, annot=True, fmt="d", cmap='Blues', square=True)
plt.xlabel("predicción")
plt.ylabel("real")
plt.title('Matriz de confusión', y=1.05, size=18)
plt.show()
print('\n')

Accuracy: 0.838
F1-binary: 0.079
Recall: 0.967


CPU times: user 1.06 s, sys: 39.5 ms, total: 1.1 s
Wall time: 791 ms


%%time

DTC= DecisionTreeClassifier(random_state=10,
                            max_depth=4,
                            min_samples_split=2,
                            class_weight= 'balanced')

DTC.fit(X_SMOTENCIHTRUS, y_SMOTENCIHTRUS)
    
pred = DTC.predict(X_val)
print("Accuracy: {:.3f}".format(accuracy_score(y_val, pred)))
print("F1-binary: {:.3f}".format(f1_score(y_val, pred, average='binary'))) 
print("Recall: {:.3f}".format(recall_score(y_val, pred))) 

matriz = confusion_matrix(y_val, pred)
sns.heatmap(matriz, annot=True, fmt="d", cmap='Blues', square=True)
plt.xlabel("predicción")
plt.ylabel("real")
plt.title('Matriz de confusión', y=1.05, size=18)
plt.show()
print('\n')

Accuracy: 0.854
F1-binary: 0.084
Recall: 0.947


CPU times: user 718 ms, sys: 37.1 ms, total: 755 ms
Wall time: 500 ms


%%time

from sklearn.tree import plot_tree

plt.subplots(figsize=(33,15))
plot_tree(DTC, feature_names = X_train.columns, 
          class_names = ['Usuario', 'Cliente'],
          filled=True, node_ids=True, fontsize=13)
plt.show()

CPU times: user 3.84 s, sys: 171 ms, total: 4.01 s
Wall time: 3.84 s


from sklearn import tree
from pydotplus import graph_from_dot_data




#import graphviz
# DOT data
dot_data = tree.export_graphviz(DTC, out_file=None, 
                                feature_names=X_train.columns,  
                                class_names=['Usuario', 'Cliente'],
                                filled=True)


#dot_data = export_graphviz(tree,
#                           feature_names=feature_names)

graph = graph_from_dot_data(dot_data)
#graph.write_png('tree.png')

# Draw graph
#graph = graphviz.Source(dot_data)


tree.export_graphviz(DTC,
                     out_file="tree.dot",
                     feature_names=X_train.columns, 
                     class_names=['Usuario', 'Cliente'],
                     rounded = True,
                     filled = True)


%%time

DTC= DecisionTreeClassifier(random_state=0,
                            max_depth=4,
                            min_samples_split=2,
                            class_weight= {0:1,1:3})

DTC.fit(X_SMOTENCIHTRUS, y_SMOTENCIHTRUS)
    
pred = DTC.predict(X_val)
print("Accuracy: {:.3f}".format(accuracy_score(y_val, pred)))
print("F1-binary: {:.3f}".format(f1_score(y_val, pred, average='binary'))) 
print("Recall: {:.3f}".format(recall_score(y_val, pred))) 

matriz = confusion_matrix(y_val, pred)
sns.heatmap(matriz, annot=True, fmt="d", cmap='Blues', square=True)
plt.xlabel("predicción")
plt.ylabel("real")
plt.title('Matriz de confusión', y=1.05, size=18)
plt.show()
print('\n')

Accuracy: 0.854
F1-binary: 0.084
Recall: 0.947


CPU times: user 636 ms, sys: 54.2 ms, total: 690 ms
Wall time: 611 ms


%%time

plt.subplots(figsize=(33,20))
plot_tree(DTC, feature_names = X_train.columns, 
          class_names = ['Usuario', 'Cliente'],
          filled=True, node_ids=True, fontsize=11)
plt.show()

CPU times: user 2.97 s, sys: 128 ms, total: 3.1 s
Wall time: 2.72 s


%%time

rfc= RandomForestClassifier(random_state=0,
                            n_estimators=50,
                            max_depth=5,
                            class_weight= 'balanced')

rfc.fit(X_SMOTENCIHTRUS, y_SMOTENCIHTRUS)
    
pred = rfc.predict(X_val)
print("Accuracy: {:.3f}".format(accuracy_score(y_val, pred)))
print("F1-binary: {:.3f}".format(f1_score(y_val, pred, average='binary'))) 
print("Recall: {:.3f}".format(recall_score(y_val, pred))) 

matriz = confusion_matrix(y_val, pred)
sns.heatmap(matriz, annot=True, fmt="d", cmap='Blues', square=True)
plt.xlabel("predicción")
plt.ylabel("real")
plt.title('Matriz de confusión', y=1.05, size=18)
plt.show()
print('\n')

Accuracy: 0.870
F1-binary: 0.094
Recall: 0.954


CPU times: user 1.27 s, sys: 38.1 ms, total: 1.31 s
Wall time: 948 ms


X_val.shape

(55156, 9)


%%time

counter = Counter(y_train)
estimate = (counter[0]/counter[1])
weights = {0:1,1:estimate}

rfc= RandomForestClassifier(random_state=0,
                            n_estimators=100,
                            max_depth=5,
                            class_weight= weights)

rfc.fit(X_SMOTENCIHTRUS, y_SMOTENCIHTRUS)
    
pred = DTC.predict(X_val)
print("Accuracy: {:.3f}".format(accuracy_score(y_val, pred)))
print("F1-binary: {:.3f}".format(f1_score(y_val, pred, average='binary'))) 
print("Recall: {:.3f}".format(recall_score(y_val, pred))) 

matriz = confusion_matrix(y_val, pred)
sns.heatmap(matriz, annot=True, fmt="d", cmap='Blues', square=True)
plt.xlabel("predicción")
plt.ylabel("real")
plt.title('Matriz de confusión', y=1.05, size=18)
plt.show()
print('\n')

Accuracy: 0.637
F1-binary: 0.038
Recall: 0.995


CPU times: user 1.17 s, sys: 36.7 ms, total: 1.2 s
Wall time: 1.29 s


%%time

counter = Counter(y_train)
estimate = (counter[0]/counter[1])*100
print(estimate)
  
xgb = XGBClassifier(scale_pos_weight=estimate, use_label_encoder=False, eval_metric='aucpr')
xgb.fit(X_SMOTENCIHTRUS, y_SMOTENCIHTRUS)
  
pred = xgb.predict(X_val)
print("Accuracy: {:.3f}".format(accuracy_score(y_val, pred)))
print("F1-binary: {:.3f}".format(f1_score(y_val, pred, average='binary'))) 
print("Recall: {:.3f}".format(recall_score(y_val, pred))) 

matriz = confusion_matrix(y_val, pred)
sns.heatmap(matriz, annot=True, fmt="d", cmap='Blues', square=True)
plt.xlabel("predicción")
plt.ylabel("real")
plt.title('Matriz de confusión', y=1.05, size=18)
plt.show()
print('\n')

13942.17130387343
Accuracy: 0.831
F1-binary: 0.077
Recall: 0.987


CPU times: user 3.44 s, sys: 181 ms, total: 3.62 s
Wall time: 1.27 s


%%time

#counter = Counter(y_SMOTENCIHTRUS)
estimate = 10000
  
xgb = XGBClassifier(scale_pos_weight=estimate, use_label_encoder=False, eval_metric='aucpr')
xgb.fit(X_SMOTENCIHTRUS, y_SMOTENCIHTRUS)
  
pred = xgb.predict(X_val)
print("Accuracy: {:.3f}".format(accuracy_score(y_val, pred)))
print("F1-binary: {:.3f}".format(f1_score(y_val, pred, average='binary'))) 
print("Recall: {:.3f}".format(recall_score(y_val, pred))) 

matriz = confusion_matrix(y_val, pred)
sns.heatmap(matriz, annot=True, fmt="d", cmap='Blues', square=True)
plt.xlabel("predicción")
plt.ylabel("real")
plt.title('Matriz de confusión', y=1.05, size=18)
plt.show()
print('\n')

Accuracy: 0.830
F1-binary: 0.076
Recall: 0.985


CPU times: user 3.14 s, sys: 259 ms, total: 3.39 s
Wall time: 1.56 s


%%time

counter = Counter(y_train)
estimate = counter[0]/counter[1]
#estimate= 10000
print(estimate)
  
xgb = XGBClassifier(scale_pos_weight=estimate, use_label_encoder=False, eval_metric='aucpr')
xgb.fit(X_train0, y_train)
  
pred = xgb.predict(X_val0)
print("Accuracy: {:.3f}".format(accuracy_score(y_val, pred)))
print("F1-binary: {:.3f}".format(f1_score(y_val, pred, average='binary'))) 
print("Recall: {:.3f}".format(recall_score(y_val, pred))) 

matriz = confusion_matrix(y_val, pred)
sns.heatmap(matriz, annot=True, fmt="d", cmap='Blues', square=True)
plt.xlabel("predicción")
plt.ylabel("real")
plt.title('Matriz de confusión', y=1.05, size=18)
plt.show()
print('\n')

139.4217130387343
Accuracy: 0.965
F1-binary: 0.243
Recall: 0.784


CPU times: user 2min 18s, sys: 2.51 s, total: 2min 20s
Wall time: 1min 19s


dill.dump_session('TFM_notebook_env.db')


%%time

print(Counter(y_train))

smotenc = SMOTENC(sampling_strategy=0.01, categorical_features=[0, 1,2], random_state=1)
iht = InstanceHardnessThreshold(random_state=1)
rus = RandomUnderSampler(random_state=1)
pipeline = Pipeline([('smotenc', smotenc), ('iht', iht), ('rus', rus)])
X_SMOTENCIHTRUS0, y_SMOTENCIHTRUS0 = pipeline.fit_resample(X_train0, y_train)

print(Counter(y_SMOTENCIHTRUS0))
print('\n')

DTC0= DecisionTreeClassifier(random_state=0,
                            max_depth=4,
                            min_samples_split=2,
                            class_weight= 'balanced')

DTC0.fit(X_SMOTENCIHTRUS0, y_SMOTENCIHTRUS0)
    
pred0 = DTC0.predict(X_val0)
print("Accuracy: {:.3f}".format(accuracy_score(y_val, pred0)))
print("F1-binary: {:.3f}".format(f1_score(y_val, pred0, average='binary'))) 
print("Recall: {:.3f}".format(recall_score(y_val, pred0))) 

matriz = confusion_matrix(y_val, pred0)
sns.heatmap(matriz, annot=True, fmt="d", cmap='Blues', square=True)
plt.xlabel("predicción")
plt.ylabel("real")
plt.title('Matriz de confusión', y=1.05, size=18)
plt.show()
print('\n')

Counter({0: 255560, 1: 1833})
Counter({0: 2555, 1: 2555})


Accuracy: 0.841
F1-binary: 0.079
Recall: 0.964


CPU times: user 3min 20s, sys: 3.01 s, total: 3min 23s
Wall time: 3min 27s


%%time

print(Counter(y_train))

smotenc = SMOTENC(sampling_strategy=0.01, categorical_features=[0, 1,2], random_state=1)
iht = InstanceHardnessThreshold(random_state=1)
rus = RandomUnderSampler(random_state=1)
pipeline = Pipeline([('smotenc', smotenc), ('iht', iht), ('rus', rus)])
X_SMOTENCIHTRUS1, y_SMOTENCIHTRUS1 = pipeline.fit_resample(X_train, y_train)

print(Counter(y_SMOTENCIHTRUS1))
print('\n')

DTC1= DecisionTreeClassifier(random_state=0,
                            max_depth=4,
                            min_samples_split=2,
                            class_weight= 'balanced')

DTC1.fit(X_SMOTENCIHTRUS1, y_SMOTENCIHTRUS1)
    
pred1 = DTC1.predict(X_val)
print("Accuracy: {:.3f}".format(accuracy_score(y_val, pred1)))
print("F1-binary: {:.3f}".format(f1_score(y_val, pred1, average='binary'))) 
print("Recall: {:.3f}".format(recall_score(y_val, pred1))) 

matriz = confusion_matrix(y_val, pred1)
sns.heatmap(matriz, annot=True, fmt="d", cmap='Blues', square=True)
plt.xlabel("predicción")
plt.ylabel("real")
plt.title('Matriz de confusión', y=1.05, size=18)
plt.show()
print('\n')

Counter({0: 255560, 1: 1833})
Counter({0: 2555, 1: 2555})


Accuracy: 0.840
F1-binary: 0.079
Recall: 0.962


CPU times: user 2min 25s, sys: 2.45 s, total: 2min 27s
Wall time: 2min 33s


%%time

print(Counter(y_train))

smotenc = SMOTENC(sampling_strategy=0.01, categorical_features=[0, 1,2], random_state=1)
iht = InstanceHardnessThreshold(random_state=1)
rus = RandomUnderSampler(sampling_strategy=0.95, random_state=1)
pipeline = Pipeline([('smotenc', smotenc), ('iht', iht), ('rus', rus)])
X_SMOTENCIHTRUS12, y_SMOTENCIHTRUS12 = pipeline.fit_resample(X_train, y_train)

print(Counter(y_SMOTENCIHTRUS12))
print('\n')

DTC12= DecisionTreeClassifier(random_state=0,
                            max_depth=4,
                            min_samples_split=2,
                            class_weight= 'balanced')

DTC12.fit(X_SMOTENCIHTRUS12, y_SMOTENCIHTRUS12)
    
pred12 = DTC12.predict(X_val)
print("Accuracy: {:.3f}".format(accuracy_score(y_val, pred12)))
print("F1-binary: {:.3f}".format(f1_score(y_val, pred12, average='binary'))) 
print("Recall: {:.3f}".format(recall_score(y_val, pred12))) 

matriz12 = confusion_matrix(y_val, pred12)
sns.heatmap(matriz12, annot=True, fmt="d", cmap='Blues', square=True)
plt.xlabel("predicción")
plt.ylabel("real")
plt.title('Matriz de confusión', y=1.05, size=18)
plt.show()
print('\n')

Counter({0: 255560, 1: 1833})
Counter({0: 2689, 1: 2555})


Accuracy: 0.839
F1-binary: 0.079
Recall: 0.962


CPU times: user 2min 31s, sys: 2.82 s, total: 2min 34s
Wall time: 2min 43s


%%time

print(Counter(y_train))

smotenc = SMOTENC(sampling_strategy=0.01, categorical_features=[0, 1,2], random_state=1)
iht = InstanceHardnessThreshold(random_state=1)
rus = RandomUnderSampler(sampling_strategy=0.9, random_state=1)
pipeline = Pipeline([('smotenc', smotenc), ('iht', iht), ('rus', rus)])
X_SMOTENCIHTRUS2, y_SMOTENCIHTRUS2 = pipeline.fit_resample(X_train, y_train)

print(Counter(y_SMOTENCIHTRUS2))
print('\n')

DTC2= DecisionTreeClassifier(random_state=0,
                            max_depth=4,
                            min_samples_split=2,
                            class_weight= 'balanced')

DTC2.fit(X_SMOTENCIHTRUS2, y_SMOTENCIHTRUS2)
    
pred2 = DTC2.predict(X_val)
print("Accuracy: {:.3f}".format(accuracy_score(y_val, pred2)))
print("F1-binary: {:.3f}".format(f1_score(y_val, pred2, average='binary'))) 
print("Recall: {:.3f}".format(recall_score(y_val, pred2))) 

matriz2 = confusion_matrix(y_val, pred2)
sns.heatmap(matriz2, annot=True, fmt="d", cmap='Blues', square=True)
plt.xlabel("predicción")
plt.ylabel("real")
plt.title('Matriz de confusión', y=1.05, size=18)
plt.show()
print('\n')

Counter({0: 255560, 1: 1833})
Counter({0: 2838, 1: 2555})


Accuracy: 0.839
F1-binary: 0.079
Recall: 0.962


CPU times: user 2min 24s, sys: 2.16 s, total: 2min 26s
Wall time: 2min 34s


%%time

print(Counter(y_train))

smotenc = SMOTENC(sampling_strategy=0.01, categorical_features=[0, 1,2], random_state=1)
iht = InstanceHardnessThreshold(random_state=1)
rus = RandomUnderSampler(sampling_strategy=0.8, random_state=1)
pipeline = Pipeline([('smotenc', smotenc), ('iht', iht), ('rus', rus)])
X_SMOTENCIHTRUS3, y_SMOTENCIHTRUS3 = pipeline.fit_resample(X_train, y_train)

print(Counter(y_SMOTENCIHTRUS3))
print('\n')

DTC3= DecisionTreeClassifier(random_state=0,
                            max_depth=4,
                            min_samples_split=2,
                            class_weight= 'balanced')

DTC3.fit(X_SMOTENCIHTRUS3, y_SMOTENCIHTRUS3)
    
pred3 = DTC3.predict(X_val)
print("Accuracy: {:.3f}".format(accuracy_score(y_val, pred3)))
print("F1-binary: {:.3f}".format(f1_score(y_val, pred3, average='binary'))) 
print("Recall: {:.3f}".format(recall_score(y_val, pred3))) 

matriz3 = confusion_matrix(y_val, pred3)
sns.heatmap(matriz3, annot=True, fmt="d", cmap='Blues', square=True)
plt.xlabel("predicción")
plt.ylabel("real")
plt.title('Matriz de confusión', y=1.05, size=18)
plt.show()
print('\n')

Counter({0: 255560, 1: 1833})
Counter({0: 3193, 1: 2555})


Accuracy: 0.841
F1-binary: 0.079
Recall: 0.962


CPU times: user 2min 23s, sys: 2.4 s, total: 2min 26s
Wall time: 2min 30s


%%time

from sklearn.tree import plot_tree

plt.subplots(figsize=(40,20))
plot_tree(DTC3, feature_names = X_train.columns, 
          class_names = ['Usuario', 'Cliente'],
          filled=True, node_ids=True, fontsize=12)
plt.show()

CPU times: user 3.58 s, sys: 150 ms, total: 3.73 s
Wall time: 3.43 s


%%time

print(Counter(y_train))

smotenc = SMOTENC(sampling_strategy=0.01, categorical_features=[0, 1,2], random_state=1)
iht = InstanceHardnessThreshold(random_state=1)
rus = RandomUnderSampler(sampling_strategy=0.8, random_state=1)
pipeline = Pipeline([('smotenc', smotenc), ('iht', iht), ('rus', rus)])
X_SMOTENCIHTRUS30, y_SMOTENCIHTRUS30 = pipeline.fit_resample(X_train0, y_train)

print(Counter(y_SMOTENCIHTRUS30))
print('\n')

DTC30= DecisionTreeClassifier(random_state=0,
                            max_depth=4,
                            min_samples_split=2,
                            class_weight= 'balanced')

DTC30.fit(X_SMOTENCIHTRUS30, y_SMOTENCIHTRUS30)
    
pred30 = DTC30.predict(X_val0)
print("Accuracy: {:.3f}".format(accuracy_score(y_val, pred30)))
print("F1-binary: {:.3f}".format(f1_score(y_val, pred30, average='binary'))) 
print("Recall: {:.3f}".format(recall_score(y_val, pred30))) 

matriz30 = confusion_matrix(y_val, pred30)
sns.heatmap(matriz30, annot=True, fmt="d", cmap='Blues', square=True)
plt.xlabel("predicción")
plt.ylabel("real")
plt.title('Matriz de confusión', y=1.05, size=18)
plt.show()
print('\n')

Counter({0: 255560, 1: 1833})
Counter({0: 3193, 1: 2555})


Accuracy: 0.840
F1-binary: 0.079
Recall: 0.964


CPU times: user 3min 17s, sys: 2.83 s, total: 3min 20s
Wall time: 3min 22s


%%time

print(Counter(y_train))

smotenc = SMOTENC(sampling_strategy=0.01, categorical_features=[0, 1,2], random_state=1)
iht = InstanceHardnessThreshold(random_state=1)
rus = RandomUnderSampler(sampling_strategy=0.7, random_state=1)
pipeline = Pipeline([('smotenc', smotenc), ('iht', iht), ('rus', rus)])
X_SMOTENCIHTRUS4, y_SMOTENCIHTRUS4 = pipeline.fit_resample(X_train, y_train)

print(Counter(y_SMOTENCIHTRUS4))
print('\n')

DTC4= DecisionTreeClassifier(random_state=0,
                            max_depth=4,
                            min_samples_split=2,
                            class_weight= 'balanced')

DTC4.fit(X_SMOTENCIHTRUS4, y_SMOTENCIHTRUS4)
    
pred4 = DTC4.predict(X_val)
print("Accuracy: {:.3f}".format(accuracy_score(y_val, pred4)))
print("F1-binary: {:.3f}".format(f1_score(y_val, pred4, average='binary'))) 
print("Recall: {:.3f}".format(recall_score(y_val, pred4))) 

matriz4 = confusion_matrix(y_val, pred4)
sns.heatmap(matriz4, annot=True, fmt="d", cmap='Blues', square=True)
plt.xlabel("predicción")
plt.ylabel("real")
plt.title('Matriz de confusión', y=1.05, size=18)
plt.show()
print('\n')

Counter({0: 255560, 1: 1833})
Counter({0: 3650, 1: 2555})


Accuracy: 0.839
F1-binary: 0.079
Recall: 0.962


CPU times: user 2min 21s, sys: 2.03 s, total: 2min 23s
Wall time: 2min 25s


%%time

print(Counter(y_train))

smotenc = SMOTENC(sampling_strategy=0.01, categorical_features=[0, 1,2], random_state=1)
iht = InstanceHardnessThreshold(random_state=1)
rus = RandomUnderSampler(sampling_strategy=0.6, random_state=1)
pipeline = Pipeline([('smotenc', smotenc), ('iht', iht), ('rus', rus)])
X_SMOTENCIHTRUS5, y_SMOTENCIHTRUS5 = pipeline.fit_resample(X_train, y_train)

print(Counter(y_SMOTENCIHTRUS5))
print('\n')

DTC5= DecisionTreeClassifier(random_state=0,
                            max_depth=4,
                            min_samples_split=2,
                            class_weight= 'balanced')

DTC5.fit(X_SMOTENCIHTRUS5, y_SMOTENCIHTRUS5)
    
pred5 = DTC5.predict(X_val)
print("Accuracy: {:.3f}".format(accuracy_score(y_val, pred5)))
print("F1-binary: {:.3f}".format(f1_score(y_val, pred5, average='binary'))) 
print("Recall: {:.3f}".format(recall_score(y_val, pred5))) 

matriz5 = confusion_matrix(y_val, pred5)
sns.heatmap(matriz5, annot=True, fmt="d", cmap='Blues', square=True)
plt.xlabel("predicción")
plt.ylabel("real")
plt.title('Matriz de confusión', y=1.05, size=18)
plt.show()
print('\n')

Counter({0: 255560, 1: 1833})
Counter({0: 4258, 1: 2555})


Accuracy: 0.848
F1-binary: 0.082
Recall: 0.949


CPU times: user 2min 21s, sys: 2.16 s, total: 2min 23s
Wall time: 2min 25s


%%time

print(Counter(y_train))

smotenc = SMOTENC(sampling_strategy=0.01, categorical_features=[0, 1,2], random_state=1)
iht = InstanceHardnessThreshold(random_state=1)
rus = RandomUnderSampler(sampling_strategy=0.5, random_state=1)
pipeline = Pipeline([('smotenc', smotenc), ('iht', iht), ('rus', rus)])
X_SMOTENCIHTRUS6, y_SMOTENCIHTRUS6 = pipeline.fit_resample(X_train, y_train)

print(Counter(y_SMOTENCIHTRUS6))
print('\n')

DTC6= DecisionTreeClassifier(random_state=0,
                            max_depth=4,
                            min_samples_split=2,
                            class_weight= 'balanced')

DTC6.fit(X_SMOTENCIHTRUS6, y_SMOTENCIHTRUS6)
    
pred6 = DTC6.predict(X_val)
print("Accuracy: {:.3f}".format(accuracy_score(y_val, pred6)))
print("F1-binary: {:.3f}".format(f1_score(y_val, pred6, average='binary'))) 
print("Recall: {:.3f}".format(recall_score(y_val, pred6))) 

matriz6 = confusion_matrix(y_val, pred6)
sns.heatmap(matriz6, annot=True, fmt="d", cmap='Blues', square=True)
plt.xlabel("predicción")
plt.ylabel("real")
plt.title('Matriz de confusión', y=1.05, size=18)
plt.show()
print('\n')

Counter({0: 255560, 1: 1833})
Counter({0: 5110, 1: 2555})


Accuracy: 0.849
F1-binary: 0.082
Recall: 0.949


CPU times: user 2min 25s, sys: 2.24 s, total: 2min 27s
Wall time: 2min 32s


%%time

print(Counter(y_train))

smotenc = SMOTENC(sampling_strategy=0.01, categorical_features=[0, 1,2], random_state=1)
iht = InstanceHardnessThreshold(random_state=1)
rus = RandomUnderSampler(sampling_strategy=0.4, random_state=1)
pipeline = Pipeline([('smotenc', smotenc), ('iht', iht), ('rus', rus)])
X_SMOTENCIHTRUS7, y_SMOTENCIHTRUS7 = pipeline.fit_resample(X_train, y_train)

print(Counter(y_SMOTENCIHTRUS7))
print('\n')

DTC7= DecisionTreeClassifier(random_state=0,
                            max_depth=4,
                            min_samples_split=2,
                            class_weight= 'balanced')

DTC7.fit(X_SMOTENCIHTRUS7, y_SMOTENCIHTRUS7)
    
pred7 = DTC7.predict(X_val)
print("Accuracy: {:.3f}".format(accuracy_score(y_val, pred7)))
print("F1-binary: {:.3f}".format(f1_score(y_val, pred7, average='binary'))) 
print("Recall: {:.3f}".format(recall_score(y_val, pred7))) 

matriz7 = confusion_matrix(y_val, pred7)
sns.heatmap(matriz7, annot=True, fmt="d", cmap='Blues', square=True)
plt.xlabel("predicción")
plt.ylabel("real")
plt.title('Matriz de confusión', y=1.05, size=18)
plt.show()
print('\n')

Counter({0: 255560, 1: 1833})
Counter({0: 6387, 1: 2555})


Accuracy: 0.854
F1-binary: 0.084
Recall: 0.947


CPU times: user 2min 19s, sys: 2.29 s, total: 2min 21s
Wall time: 2min 23s


%%time

print(Counter(y_train))

smotenc = SMOTENC(sampling_strategy=0.01, categorical_features=[0, 1,2], random_state=1)
iht = InstanceHardnessThreshold(random_state=1)
rus = RandomUnderSampler(sampling_strategy=0.3, random_state=1)
pipeline = Pipeline([('smotenc', smotenc), ('iht', iht), ('rus', rus)])
X_SMOTENCIHTRUS8, y_SMOTENCIHTRUS8 = pipeline.fit_resample(X_train, y_train)

print(Counter(y_SMOTENCIHTRUS8))
print('\n')

DTC8= DecisionTreeClassifier(random_state=0,
                            max_depth=4,
                            min_samples_split=2,
                            class_weight= 'balanced')

DTC8.fit(X_SMOTENCIHTRUS8, y_SMOTENCIHTRUS8)
    
pred8 = DTC8.predict(X_val)
print("Accuracy: {:.3f}".format(accuracy_score(y_val, pred8)))
print("F1-binary: {:.3f}".format(f1_score(y_val, pred8, average='binary'))) 
print("Recall: {:.3f}".format(recall_score(y_val, pred8))) 

matriz8 = confusion_matrix(y_val, pred8)
sns.heatmap(matriz8, annot=True, fmt="d", cmap='Blues', square=True)
plt.xlabel("predicción")
plt.ylabel("real")
plt.title('Matriz de confusión', y=1.05, size=18)
plt.show()
print('\n')

Counter({0: 255560, 1: 1833})
Counter({0: 8516, 1: 2555})


Accuracy: 0.840
F1-binary: 0.079
Recall: 0.962


CPU times: user 2min 19s, sys: 2.01 s, total: 2min 21s
Wall time: 2min 25s


%%time

print(Counter(y_train))

smotenc = SMOTENC(sampling_strategy=0.01, categorical_features=[0, 1,2], random_state=1)
iht = InstanceHardnessThreshold(random_state=1)
rus = RandomUnderSampler(sampling_strategy=0.2, random_state=1)
pipeline = Pipeline([('smotenc', smotenc), ('iht', iht), ('rus', rus)])
X_SMOTENCIHTRUS9, y_SMOTENCIHTRUS9 = pipeline.fit_resample(X_train, y_train)

print(Counter(y_SMOTENCIHTRUS9))
print('\n')

DTC9= DecisionTreeClassifier(random_state=0,
                            max_depth=4,
                            min_samples_split=2,
                            class_weight= 'balanced')

DTC9.fit(X_SMOTENCIHTRUS8, y_SMOTENCIHTRUS8)
    
pred9 = DTC9.predict(X_val)
print("Accuracy: {:.3f}".format(accuracy_score(y_val, pred9)))
print("F1-binary: {:.3f}".format(f1_score(y_val, pred9, average='binary'))) 
print("Recall: {:.3f}".format(recall_score(y_val, pred9))) 

matriz9 = confusion_matrix(y_val, pred9)
sns.heatmap(matriz9, annot=True, fmt="d", cmap='Blues', square=True)
plt.xlabel("predicción")
plt.ylabel("real")
plt.title('Matriz de confusión', y=1.05, size=18)
plt.show()
print('\n')

Counter({0: 255560, 1: 1833})
Counter({0: 12775, 1: 2555})


Accuracy: 0.840
F1-binary: 0.079
Recall: 0.962


CPU times: user 2min 22s, sys: 2.27 s, total: 2min 24s
Wall time: 2min 27s


%%time

print(Counter(y_train))

smotenc = SMOTENC(sampling_strategy=0.01, categorical_features=[0, 1,2], random_state=1)
iht = InstanceHardnessThreshold(random_state=1)
rus = RandomUnderSampler(sampling_strategy=0.1, random_state=1)
pipeline = Pipeline([('smotenc', smotenc), ('iht', iht), ('rus', rus)])
X_SMOTENCIHTRUS9, y_SMOTENCIHTRUS9 = pipeline.fit_resample(X_train, y_train)

print(Counter(y_SMOTENCIHTRUS9))
print('\n')

DTC9= DecisionTreeClassifier(random_state=0,
                            max_depth=4,
                            min_samples_split=2,
                            class_weight= 'balanced')

DTC9.fit(X_SMOTENCIHTRUS8, y_SMOTENCIHTRUS8)
    
pred9 = DTC9.predict(X_val)
print("Accuracy: {:.3f}".format(accuracy_score(y_val, pred9)))
print("F1-binary: {:.3f}".format(f1_score(y_val, pred9, average='binary'))) 
print("Recall: {:.3f}".format(recall_score(y_val, pred9))) 

matriz9 = confusion_matrix(y_val, pred9)
sns.heatmap(matriz9, annot=True, fmt="d", cmap='Blues', square=True)
plt.xlabel("predicción")
plt.ylabel("real")
plt.title('Matriz de confusión', y=1.05, size=18)
plt.show()
print('\n')

Counter({0: 255560, 1: 1833})
Counter({0: 25550, 1: 2555})


Accuracy: 0.840
F1-binary: 0.079
Recall: 0.962


CPU times: user 2min 22s, sys: 2.01 s, total: 2min 24s
Wall time: 2min 26s


%%time

print(Counter(y_train))

smotenc = SMOTENC(sampling_strategy=0.7, categorical_features=[0, 1,2], random_state=1)
iht = InstanceHardnessThreshold(random_state=1)
#rus = RandomUnderSampler(sampling_strategy=0.1, random_state=1)
pipeline = Pipeline([
    ('smotenc', smotenc), 
    ('iht', iht), 
 #   ('rus', rus)
    ])
X_SMOTENCIHTRUS9, y_SMOTENCIHTRUS9 = pipeline.fit_resample(X_train, y_train)

print(Counter(y_SMOTENCIHTRUS9))
print('\n')

DTC9= DecisionTreeClassifier(random_state=0,
                            max_depth=4,
                            min_samples_split=2,
                            class_weight= 'balanced')

DTC9.fit(X_SMOTENCIHTRUS8, y_SMOTENCIHTRUS8)
    
pred9 = DTC9.predict(X_val)
print("Accuracy: {:.3f}".format(accuracy_score(y_val, pred9)))
print("F1-binary: {:.3f}".format(f1_score(y_val, pred9, average='binary'))) 
print("Recall: {:.3f}".format(recall_score(y_val, pred9))) 

matriz9 = confusion_matrix(y_val, pred9)
sns.heatmap(matriz9, annot=True, fmt="d", cmap='Blues', square=True)
plt.xlabel("predicción")
plt.ylabel("real")
plt.title('Matriz de confusión', y=1.05, size=18)
plt.show()
print('\n')

Counter({0: 255560, 1: 1833})
Counter({0: 219381, 1: 178892})


Accuracy: 0.840
F1-binary: 0.079
Recall: 0.962


CPU times: user 5min 14s, sys: 6.57 s, total: 5min 21s
Wall time: 5min 26s


%%time

print(Counter(y_train))

smotenc = SMOTENC(sampling_strategy=1, categorical_features=[0, 1,2], random_state=1)
#iht = InstanceHardnessThreshold(random_state=1)
#rus = RandomUnderSampler(sampling_strategy=0.1, random_state=1)
pipeline = Pipeline([
    ('smotenc', smotenc), 
#    ('iht', iht), 
#    ('rus', rus)
    ])
X_SMOTENCIHTRUS9, y_SMOTENCIHTRUS9 = pipeline.fit_resample(X_train, y_train)

print(Counter(y_SMOTENCIHTRUS9))
print('\n')

DTC9= DecisionTreeClassifier(random_state=0,
                            max_depth=4,
                            min_samples_split=2,
                            class_weight= 'balanced')

DTC9.fit(X_SMOTENCIHTRUS8, y_SMOTENCIHTRUS8)
    
pred9 = DTC9.predict(X_val)
print("Accuracy: {:.3f}".format(accuracy_score(y_val, pred9)))
print("F1-binary: {:.3f}".format(f1_score(y_val, pred9, average='binary'))) 
print("Recall: {:.3f}".format(recall_score(y_val, pred9))) 

matriz9 = confusion_matrix(y_val, pred9)
sns.heatmap(matriz9, annot=True, fmt="d", cmap='Blues', square=True)
plt.xlabel("predicción")
plt.ylabel("real")
plt.title('Matriz de confusión', y=1.05, size=18)
plt.show()
print('\n')

Counter({0: 255560, 1: 1833})
Counter({0: 255560, 1: 255560})


Accuracy: 0.840
F1-binary: 0.079
Recall: 0.962


CPU times: user 32.1 s, sys: 1.17 s, total: 33.3 s
Wall time: 33.7 s


%%time

print(Counter(y_train))

#smotenc = SMOTENC(sampling_strategy=0.01, categorical_features=[0, 1,2], random_state=1)
iht = InstanceHardnessThreshold(random_state=1)
#rus = RandomUnderSampler(sampling_strategy=0.1, random_state=1)
pipeline = Pipeline([
#    ('smotenc', smotenc), 
    ('iht', iht), 
#    ('rus', rus)
    ])
X_SMOTENCIHTRUS9, y_SMOTENCIHTRUS9 = pipeline.fit_resample(X_train, y_train)

print(Counter(y_SMOTENCIHTRUS9))
print('\n')

DTC9= DecisionTreeClassifier(random_state=0,
                            max_depth=4,
                            min_samples_split=2,
                            class_weight= 'balanced')

DTC9.fit(X_SMOTENCIHTRUS8, y_SMOTENCIHTRUS8)
    
pred9 = DTC9.predict(X_val)
print("Accuracy: {:.3f}".format(accuracy_score(y_val, pred9)))
print("F1-binary: {:.3f}".format(f1_score(y_val, pred9, average='binary'))) 
print("Recall: {:.3f}".format(recall_score(y_val, pred9))) 

matriz9 = confusion_matrix(y_val, pred9)
sns.heatmap(matriz9, annot=True, fmt="d", cmap='Blues', square=True)
plt.xlabel("predicción")
plt.ylabel("real")
plt.title('Matriz de confusión', y=1.05, size=18)
plt.show()
print('\n')

Counter({0: 255560, 1: 1833})
Counter({0: 237525, 1: 1833})


Accuracy: 0.840
F1-binary: 0.079
Recall: 0.962


CPU times: user 2min 5s, sys: 1.62 s, total: 2min 6s
Wall time: 2min 8s


%%time

print(Counter(y_train))

#smotenc = SMOTENC(sampling_strategy=0.01, categorical_features=[0, 1,2], random_state=1)
#iht = InstanceHardnessThreshold(random_state=1)
rus = RandomUnderSampler(sampling_strategy=0.01, random_state=1)
pipeline = Pipeline([
#    ('smotenc', smotenc), 
#    ('iht', iht), 
    ('rus', rus)
    ])
X_SMOTENCIHTRUS9, y_SMOTENCIHTRUS9 = pipeline.fit_resample(X_train, y_train)

print(Counter(y_SMOTENCIHTRUS9))
print('\n')

DTC9= DecisionTreeClassifier(random_state=0,
                            max_depth=4,
                            min_samples_split=2,
                            class_weight= 'balanced')

DTC9.fit(X_SMOTENCIHTRUS8, y_SMOTENCIHTRUS8)
    
pred9 = DTC9.predict(X_val)
print("Accuracy: {:.3f}".format(accuracy_score(y_val, pred9)))
print("F1-binary: {:.3f}".format(f1_score(y_val, pred9, average='binary'))) 
print("Recall: {:.3f}".format(recall_score(y_val, pred9))) 

matriz9 = confusion_matrix(y_val, pred9)
sns.heatmap(matriz9, annot=True, fmt="d", cmap='Blues', square=True)
plt.xlabel("predicción")
plt.ylabel("real")
plt.title('Matriz de confusión', y=1.05, size=18)
plt.show()
print('\n')

Counter({0: 255560, 1: 1833})
Counter({0: 183300, 1: 1833})


Accuracy: 0.840
F1-binary: 0.079
Recall: 0.962


CPU times: user 1.09 s, sys: 77.2 ms, total: 1.17 s
Wall time: 840 ms


%%time

#print(Counter(y_train))

#smotenc = SMOTENC(sampling_strategy=0.01, categorical_features=[0, 1,2], random_state=1)
#iht = InstanceHardnessThreshold(random_state=1)
#rus = RandomUnderSampler(sampling_strategy=0.1, random_state=1)
#pipeline = Pipeline([
#    ('smotenc', smotenc), 
#    ('iht', iht), 
#    ('rus', rus)
#    ])
#X_SMOTENCIHTRUS9, y_SMOTENCIHTRUS9 = pipeline.fit_resample(X_train, y_train)

#print(Counter(y_SMOTENCIHTRUS9))
#print('\n')

DTC9= DecisionTreeClassifier(random_state=0,
                            max_depth=4,
                            min_samples_split=2,
                            class_weight= 'balanced')

#DTC9.fit(X_SMOTENCIHTRUS8, y_SMOTENCIHTRUS8)
DTC9.fit(X_train, y_train)
    
pred9 = DTC9.predict(X_val)
print("Accuracy: {:.3f}".format(accuracy_score(y_val, pred9)))
print("F1-binary: {:.3f}".format(f1_score(y_val, pred9, average='binary'))) 
print("Recall: {:.3f}".format(recall_score(y_val, pred9))) 

matriz9 = confusion_matrix(y_val, pred9)
sns.heatmap(matriz9, annot=True, fmt="d", cmap='Blues', square=True)
plt.xlabel("predicción")
plt.ylabel("real")
plt.title('Matriz de confusión', y=1.05, size=18)
plt.show()
print('\n')

Accuracy: 0.856
F1-binary: 0.086
Recall: 0.944


CPU times: user 1.39 s, sys: 77.6 ms, total: 1.47 s
Wall time: 1.19 s

	IDUSUARIO	TIPOUSUARIO	FEC_REGISTRO	CANAL_REGISTRO	TIPOEMAIL	BONDAD_EMAIL	USU_TELF	IPCASOS	IP_Country	IP_Area	USU_TIPO	USU_TAMANIO	USU_CIIU	USU_ESTADO	USU_DEPARTAMENTO	FEC_CLIENTE	FEC_ALTA
0	8107310.0	PF	22/10/2019 0:00:00	3.0	yahoo.com	0	233XXXXX	1	Colombia	America	NaN	NaN	NaN	NaN	NaN	NaN	NaN
1	7784565.0	PJ	14/5/2019 0:00:00	3.0	gmail.com	20	633XXXXX	1	Colombia	America	SOCIEDAD COMERCIAL/INDUSTRIAL	PQ	N7820	ACTIVA	QUINDIO	NaN	NaN
2	7718778.0	PJ	9/4/2019 0:00:00	7.0	hotmail.com	20	533XXXXX	1	Colombia	America	SOCIEDAD COMERCIAL/INDUSTRIAL	MC	G4774	ACTIVA	ATLANTICO	NaN	NaN
3	7952765.0	PX	12/8/2019 0:00:00	3.0	uqvirtual.edu.co	20	633XXXXX	1	Colombia	America	NaN	NaN	NaN	NaN	NaN	NaN	NaN
4	7855424.0	PJ	21/6/2019 0:00:00	7.0	hotmail.com	20	533XXXXX	1	Colombia	America	EMPRESARIO INDIVIDUAL	MC	N8299	CANCELACIÓN	ATLANTICO	NaN	NaN

	IDCONSUMO	IDUSUARIO	IDPRODUCTO	DESCPRODUCTO	FECHACONSUMO	EMPCONSUL_ID	EMPCONSUL_CIIU	EMPCONSUL_PROV	EMPCONSUL_EST
0	40057306.0	6868835.0	144920.0	Perfil Promocional	1/1/2018 0:00:00	986515	H4921	ATLANTICO	ACTIVA
1	40057715.0	6868838.0	144920.0	Perfil Promocional	1/1/2018 0:00:00	102614	N8121	BOGOTA	ACTIVA
2	40057758.0	6868841.0	144920.0	Perfil Promocional	1/1/2018 0:00:00	915924	S9609	BOGOTA	ACTIVA
3	40057765.0	6868844.0	144920.0	Perfil Promocional	1/1/2018 0:00:00	3830723	G4642	BOGOTA	ACTIVA
4	40057767.0	6868841.0	144917.0	Ficha Básica	1/1/2018 0:00:00	2971771	P8544	BOGOTA	ACTIVA

	IDCONSUMO	IDUSUARIO	FEC_CLIENTE	IDPRODUCTO	DESCPRODUCTO	FECHACONSUMO	EMPCONSUL_ID	EMPCONSUL_CIIU	EMPCONSUL_PROV	EMPCONSUL_EST
382	40068583	6869752	2020-09-30	144917	Ficha Básica	2018-02-01	3597384	H5221	BOGOTA	ACTIVA
384	40068664	6869761	2018-02-01	144920	Perfil Promocional	2018-02-01	2857218	C2819	BOGOTA	ACTIVA
394	40069063	6869761	2018-02-01	144917	Ficha Básica	2018-02-01	983852	G4754	ATLANTICO	LEY DE INSOLVENCIA (REORGANIZACION EMPRESARIAL)
395	40069077	6869761	2018-02-01	144917	Ficha Básica	2018-02-01	2857218	C2819	BOGOTA	ACTIVA
397	40069177	6869802	2018-03-01	144917	Ficha Básica	2018-02-01	970999	J6202	BOGOTA	ACTIVA

	IDCONSUMO	IDUSUARIO	FEC_CLIENTE	IDPRODUCTO	DESCPRODUCTO	FECHACONSUMO	EMPCONSUL_ID	EMPCONSUL_CIIU	EMPCONSUL_PROV	EMPCONSUL_EST
2062	40105110	6873624	2018-04-30	144920	Perfil Promocional	2018-05-01	374863	C2511	VALLE	ACTIVA
2983	40121009	6873020	2018-05-02	144920	Perfil Promocional	2018-08-01	1576337	J6209	BOGOTA	ACTIVA
3624	40130890	6876844	2018-01-15	144920	Perfil Promocional	2018-09-01	3455346	G4651	CUNDINAMARCA	ACTIVA
4605	40148910	6878961	2018-06-02	144920	Perfil Promocional	2018-10-01	916567	K6499	BOGOTA	ACTIVA
5602	40166323	6876844	2018-01-15	144917	Ficha Básica	2018-11-01	915197	C3110	CUNDINAMARCA	LEY DE INSOLVENCIA (REORGANIZACION EMPRESARIAL)
...	...	...	...	...	...	...	...	...	...	...
794677	58741666	7682120	2020-03-09	144917	Ficha Básica	2020-08-23	3415683	C1812	ANTIOQUIA	ACTIVA
794678	58745226	7682120	2020-03-09	144917	Ficha Básica	2020-08-23	3415683	C1812	ANTIOQUIA	ACTIVA
794980	59123301	7952617	2020-10-11	144917	Ficha Básica	2020-11-09	1232072	C1811	VALLE	ACTIVA
794983	59128242	7952617	2020-10-11	144917	Ficha Básica	2020-11-09	584373	M6920	VALLE	CANCELACION
794984	59128415	7952617	2020-10-11	144917	Ficha Básica	2020-11-09	3384462	G4690	VALLE	ACTIVA

	IDUSUARIO	primer_consumo	último_consumo
0	6868835	2018-01-01	2018-01-01
1	6868838	2018-01-01	2018-01-01
2	6868841	2018-01-01	2018-01-01
3	6868844	2018-01-01	2018-01-01
4	6868847	2018-01-01	2018-01-01
...	...	...	...
367633	8250344	2019-12-31	2019-12-31
367634	8250350	2019-12-31	2019-12-31
367635	8250356	2019-12-31	2019-12-31
367636	8250359	2019-12-31	2019-12-31
367637	8250362	2019-12-31	2019-12-31

DESCPRODUCTO	Ficha Básica	Perfil Promocional
IDUSUARIO
6868835	0	1
7718639	0	1
7718636	0	1
7718630	0	1
7718624	0	1
...	...	...
6977819	1085	0
7805198	1659	1
7542443	2062	0
7745584	2342	1
7943229	3206	1

DESCPRODUCTO	Ficha Básica	Perfil Promocional
IDUSUARIO
7592780	1	0
7265523	1	0
7710896	1	0
7604300	1	0
7710905	1	0
...	...	...
7192703	6	5
8097673	5	5
7192712	5	5
7989233	4	5
7225846	6	5

	IDUSUARIO	TIPOUSUARIO	FEC_REGISTRO	CANAL_REGISTRO	TIPOEMAIL	BONDAD_EMAIL	USU_TELF	IPCASOS	...	USU_TAMANIO	USU_CIIU	USU_ESTADO	USU_DEPARTAMENTO	FEC_CLIENTE	FEC_ALTA	Perfil Promocional	primer_consumo	último_consumo
0	8107310	PF	2019-10-22	3.0	yahoo.com	0	233XXXXX	1	...	NaN	NaN	NaN	NaN	NaT	NaT	1.0	2019-10-22	2019-10-22
1	7784565	PJ	2019-05-14	3.0	gmail.com	20	633XXXXX	1	...	PQ	N7820	ACTIVA	QUINDIO	NaT	NaT	1.0	2019-05-14	2019-05-14
2	7718778	PJ	2019-09-04	7.0	hotmail.com	20	533XXXXX	1	...	MC	G4774	ACTIVA	ATLANTICO	NaT	NaT	1.0	2019-09-04	2019-09-04
3	7952765	PX	2019-12-08	3.0	uqvirtual.edu.co	20	633XXXXX	1	...	NaN	NaN	NaN	NaN	NaT	NaT	1.0	2019-12-08	2019-12-08
4	7855424	PJ	2019-06-21	7.0	hotmail.com	20	533XXXXX	1	...	MC	N8299	CANCELACIÓN	ATLANTICO	NaT	NaT	1.0	2019-06-21	2019-06-21

	IDUSUARIO	FECHALOGIN
0	7339406.0	31/8/2018 0:00:00
1	7339406.0	4/9/2018 0:00:00
2	7339406.0	4/9/2018 0:00:00
3	7966293.0	10/10/2019 0:00:00
4	7694472.0	28/3/2019 0:00:00

	IDUSUARIO	FECHALOGIN	FEC_CLIENTE
0	7339406	2018-08-31	2018-04-09
1	7339406	2018-04-09	2018-04-09
2	7339406	2018-04-09	2018-04-09
2275	8042554	2019-09-23	2019-09-23
2276	8042554	2019-09-23	2019-09-23

	IDUSUARIO	primer_login	último_login
0	6868713	2017-12-31	2017-12-31
1	6868722	2018-01-15	2018-01-15
2	6868736	2018-02-01	2019-07-29
3	6868739	2018-05-24	2021-01-23
4	6868742	2017-12-31	2017-12-31
...	...	...	...
197124	8250329	2020-01-07	2020-10-15
197125	8250335	2020-12-11	2020-12-11
197126	8250338	2020-10-13	2020-10-13
197127	8250341	2019-12-31	2020-02-01
197128	8250362	2020-01-03	2021-01-26

	IDUSUARIO	FECHA_SESION	SESIONES	FEC_CLIENTE
1209	6873020	2018-08-01	1	2018-05-02
1399	6873624	2018-05-01	3	2018-04-30
2377	6876844	2018-09-01	3	2018-01-15
2378	6876844	2018-11-01	6	2018-01-15
3050	6878961	2018-10-01	5	2018-06-02
...	...	...	...	...
386172	8176531	2019-11-26	8	2019-01-12
386174	8176531	2019-11-21	14	2019-01-12
390871	8191801	2019-11-28	5	2019-01-12
392284	8196380	2019-11-30	5	2019-01-12
392336	8196562	2019-11-30	16	2019-02-12

	FEC_REGISTRO	FEC_CLIENTE
39887	2019-01-16	2021-04-01
317034	2019-02-14	2021-01-22
205286	2019-04-04	2021-01-22
337114	2019-04-17	2021-01-22
92079	2018-06-08	2021-01-22
...	...	...
367700	2019-06-11	NaT
367701	2019-08-11	NaT
367702	2019-12-11	NaT
367703	2019-11-18	NaT
367704	2019-05-12	NaT

	días
count	1057
mean	225 days 22:11:00.737937560
std	233 days 00:14:34.337356052
min	1 days 00:00:00
25%	31 days 00:00:00
50%	147 days 00:00:00
75%	371 days 00:00:00
max	991 days 00:00:00

	días
count	2613
mean	84 days 23:02:08.128587830
std	190 days 22:34:33.031432278
min	-322 days +00:00:00
25%	0 days 00:00:00
50%	0 days 00:00:00
75%	77 days 00:00:00
max	991 days 00:00:00

	IDUSUARIO	FECHALOGIN	FEC_CLIENTE
0	7339406	2018-08-31	2018-04-09
3608	7535569	2019-10-01	2019-04-12
3609	7535569	2019-10-01	2019-04-12
3610	7535569	2019-12-03	2019-04-12
3613	7535569	2019-10-01	2019-04-12
...	...	...	...
614807	7072056	2019-09-05	2019-06-18
615127	7953035	2019-12-08	2019-08-14
615128	7953035	2019-12-08	2019-08-14
615137	7371523	2018-09-18	2018-07-10
617724	7417448	2019-09-03	2019-06-28

	Nº nulos	% nulos
IDUSUARIO	0	0.0%
TIPOUSUARIO	0	0.0%
FEC_REGISTRO	0	0.0%
CANAL_REGISTRO	7534	2.05%
IND_CLIENTE	0	0.0%
IND_ALTA	0	0.0%
TIPOEMAIL	0	0.0%
BONDAD_EMAIL	0	0.0%
USU_TELF	66	0.02%
IPCASOS	0	0.0%
IP_Country	21764	5.92%
IP_Area	21764	5.92%
USU_TIPO	283591	77.12%
USU_TAMANIO	283589	77.12%
USU_CIIU	283589	77.12%
USU_ESTADO	283589	77.12%
USU_DEPARTAMENTO	277197	75.39%
FEC_CLIENTE	365092	99.29%
FEC_ALTA	363994	98.99%
Ficha Básica	0	0.0%
Perfil Promocional	0	0.0%
primer_consumo	67	0.02%
último_consumo	67	0.02%
N_logins	0	0.0%
primer_login	170625	46.4%
último_login	170625	46.4%
N_sesiones	0	0.0%
primera_sesión	3	0.0%
última_sesión	3	0.0%
díasEntreConsumos	0	0.0%
díasEntreLogins	0	0.0%
díasEntreSesiones	0	0.0%

	counts	percentage
PF	265760	72.28%
PJ	89824	24.43%
PX	12121	3.3%

		counts	%
IND_CLIENTE	TIPOUSUARIO
0	PF	264584	72.47%
	PJ	88521	24.25%
	PX	11981	3.28%
1	PJ	1303	49.75%
	PF	1176	44.9%
	PX	140	5.35%

	counts	%
canales_populares	240725	65.47%
canales_especializados	66628	18.12%
SEM	36461	9.92%
SE0	16357	4.45%
otros_canales	7534	2.05%

	counts	percentage
gmail.com	159897	43.49%
hotmail.com	121607	33.07%
yahoo.com	6503	1.77%
yahoo.es	5239	1.42%
outlook.com	4884	1.33%
hotmail.es	3638	0.99%
yopmail.com	3366	0.92%
misena.edu.co	2792	0.76%
outlook.es	1764	0.48%
live.com	1062	0.29%
cool.fr.nf	931	0.25%
uniminuto.edu.co	744	0.2%
unal.edu.co	644	0.18%
gamil.com	559	0.15%
courriel.fr.nf	557	0.15%

	counts	percentage
gmail.com	839	32.04%
hotmail.com	694	26.5%
yahoo.com	82	3.13%
outlook.com	37	1.41%
yahoo.es	32	1.22%
icloud.com	12	0.46%
une.net.co	8	0.31%
hotmail.es	8	0.31%
msn.com	7	0.27%
uniandes.edu.co	7	0.27%
live.com	6	0.23%
yahoo.com.mx	4	0.15%
yahoo.com.co	4	0.15%
outlook.es	4	0.15%
me.com	4	0.15%

		counts	percentage
TIPOEMAIL	IND_CLIENTE
email_colombiano	0	14944	97.74%
email_colombiano	1	346	2.26%
email_educativo	0	11007	99.54%
email_educativo	1	51	0.46%
email_español	0	11240	99.51%
email_español	1	55	0.49%
email_microsoft	0	6399	99.32%
email_microsoft	1	44	0.68%
email_minoritario	0	11096	99.27%
email_minoritario	1	82	0.73%
email_muy_minoritario	0	16823	97.62%
email_muy_minoritario	1	411	2.38%
gmail.com	0	159058	99.48%
gmail.com	1	839	0.52%
hotmail.com	0	120913	99.43%
hotmail.com	1	694	0.57%
otros_email	0	7185	99.79%
otros_email	1	15	0.21%
yahoo.com	0	6421	98.74%
yahoo.com	1	82	1.26%

		counts	percentage
BONDAD_EMAIL	IND_CLIENTE
-20	0	12016	99.71%
-20	1	35	0.29%
-10	0	16808	99.9%
-10	1	17	0.1%
0	0	54000	99.88%
0	1	67	0.12%
1	0	4917	99.45%
1	1	27	0.55%
9	0	4828	99.38%
9	1	30	0.62%
20	0	272517	99.11%
20	1	2443	0.89%

		counts	percentage
IND_CLIENTE	BONDAD_EMAIL
0	email_OK	272517	74.64%
	email_Rojo	54000	14.79%
	email_sin_dominio	16808	4.6%
	No_email	12016	3.29%
	SPAM	4917	1.35%
	email_Naranja	4828	1.32%
1	email_OK	2443	93.28%
	email_Rojo	67	2.56%
	No_email	35	1.34%
	email_Naranja	30	1.15%
	SPAM	27	1.03%
	email_sin_dominio	17	0.65%

	Columna	Chi2	Hipótesis
0	TIPOUSUARIO	999.517	Se rechaza la hipótesis nula
1	CANAL_REGISTRO	752.7108	Se rechaza la hipótesis nula
2	TIPOEMAIL	1387.2722	Se rechaza la hipótesis nula
3	BONDAD_EMAIL	507.1491	Se rechaza la hipótesis nula

	Columna	fvalor	p-valor	Hipótesis
0	IPCASOS	104.815239	0.0	Se rechaza la hipótesis nula
1	Ficha Básica	265.662249	0.0	Se rechaza la hipótesis nula
2	Perfil Promocional	74.975647	0.0	Se rechaza la hipótesis nula
3	N_logins	2362.100712	0.0	Se rechaza la hipótesis nula
4	N_sesiones	1040.505436	0.0	Se rechaza la hipótesis nula
5	díasEntreConsumos	1880.656099	0.0	Se rechaza la hipótesis nula
6	díasEntreLogins	19.560644	0.00001	Se rechaza la hipótesis nula
7	díasEntreSesiones	1807.612849	0.0	Se rechaza la hipótesis nula

	No_email	PF	PJ	PX	SE0	SEM	SPAM	canales_especializados	canales_populares	email_Naranja	...	hotmail.com	otros_canales	otros_email	yahoo.com	IPCASOS	Perfil Promocional	N_logins	N_sesiones	díasEntreConsumos	díasEntreLogins
0	0.0	1.0	0.0	0.0	0.0	1.0	0.0	0.0	1.0	0.0	...	0.0	1.0	0.0	0.0	0	0	2	4	0	0
1	1.0	0.0	0.0	0.0	0.0	1.0	0.0	0.0	0.0	0.0	...	0.0	0.0	1.0	0.0	1	1	1	3	0	0
2	1.0	0.0	0.0	0.0	0.0	0.0	1.0	0.0	0.0	0.0	...	0.0	1.0	0.0	0.0	1	1	1	6	86	0
3	1.0	0.0	0.0	0.0	0.0	0.0	1.0	0.0	0.0	0.0	...	0.0	1.0	0.0	0.0	5	1	2	3	0	0
4	1.0	0.0	0.0	0.0	1.0	0.0	0.0	0.0	0.0	0.0	...	0.0	0.0	1.0	0.0	3739	2	7	9	0	0
...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...
257388	1.0	0.0	0.0	0.0	0.0	0.0	1.0	0.0	0.0	0.0	...	0.0	1.0	0.0	0.0	1	1	0	3	0	0
257389	1.0	0.0	0.0	1.0	0.0	0.0	0.0	0.0	0.0	0.0	...	0.0	0.0	1.0	0.0	3612	2	0	7	0	0
257390	1.0	0.0	0.0	0.0	0.0	1.0	0.0	0.0	0.0	0.0	...	0.0	1.0	0.0	0.0	3	1	0	3	0	0
257391	0.0	1.0	0.0	0.0	0.0	0.0	1.0	0.0	0.0	0.0	...	0.0	1.0	0.0	0.0	1	1	0	3	0	0
257392	1.0	0.0	0.0	0.0	1.0	0.0	0.0	0.0	0.0	0.0	...	0.0	0.0	0.0	1.0	3	1	1	4	0	0

Trabajo de Final del Máster universitario en Ciencia de Datos¶

Predicción de potenciales compradores en un ecommerce¶

Importo las libresrías que voy a usar¶

Carga de los datasets¶

USUARIOS¶

CONSUMOS¶

LOGINS¶

SESIONES¶

Exploración y preparación de los datos¶

Desbalanceo¶

Fechas¶

Valores perdidos¶

Eliminación campos que no interesan¶

Informe exploratorio descriptivo¶

Variables Categóricas¶

TIPOUSUARIO¶

CANAL_REGISTRO¶

TIPOEMAIL¶

BONDAD_EMAIL¶

Selección de características¶

Variables Categóricas¶

CHI2¶

Variables numéricas¶

Correlaciones¶

ANOVA¶

Separación en entrenamiento y test¶

Preprocesamiento¶

ONE HOT ENCODING¶

Recursive Feature Elimination¶

Data Sampling sobre train¶

Oversample¶

Random Oversampling¶

SMOTE¶

SMOTENC¶

Borderline-SMOTE¶

Borderline-SMOTE SVM¶

Adaptive Synthetic Sampling (ADASYN)¶

Undersample¶

Random Undersample¶

Near Miss Undersampling¶

ClusterCentroids¶

Condensed Nearest Neighbor Rule Undersampling¶

Tomek Links for Undersampling¶

Edited Nearest Neighbors Rule for Undersampling¶

Repeated Edited Nearest Neighbours¶

AllKNN¶

Instance Hardness Threshold¶

One-Sided Selection for Undersampling¶

Neighbourhood Cleaning Rule¶

Resample¶

SMOTE and Tomek Links Undersampling¶

SMOTE and Edited Nearest Neighbors Undersampling¶

Modelos de clasificación¶

Modelo base (baseline model)¶

Regresión logística¶

Naive-Bayes¶

K vecinos más próximos¶

Árbol de decisión¶

Máquina de soporte vectorial¶

XGBoost¶

XGBoost con pesos balanceados¶

Random Forest¶

RandomForestClassifier con pesos balanceados¶

Elección de dataset de Train¶

Oversampling¶

Undersampling¶

Resample¶

Combinación de técnicas de resampling que mejor resultado han dado¶

Selección de dataset resampleado¶

Selección del algortimo final¶

Ajuste del algoritmo seleccionado¶

Ajuste del algoritmo Random Forest¶

Ajuste del algoritmo Árbol de decisión¶

Algoritmo final y validación¶

Prueba final¶